A Comprehensive Review: Study of Artificial Intelligence Optimization Technique Applications in a Hybrid Microgrid at Times of Fault Outbreaks

Abstract

The use of fossil-fueled power stations to generate electricity has had a damaging effect over the years, necessitating the need for alternative energy sources. Microgrids consisting of renewable energy source concepts have gained a lot of consideration in recent years as an alternative because they use advances in information and communication technology (ICT) to increase the quality and efficiency of services and distributed energy resources (DERs), which are environmentally friendly. Nevertheless, microgrids are constrained by the outbreaks of faults, which have an impact on their performance and necessitate dynamic energy management and optimization strategies. The application of artificial intelligence (AI) is gaining momentum as a vital key at this point. This study focuses on a comprehensive review of applications of artificial intelligence strategies on hybrid renewable microgrids for optimization, power quality enhancement, and analyses of fault outbreaks in microgrids. The use of techniques such as machine learning (ML), genetic algorithms (GA), artificial neural networks (ANN), fuzzy logic (FL), particle swarm optimization (PSO), heuristic optimization, artificial bee colony (ABC), and others is reviewed for various microgrid strategies such as regression and classification in this study. Applications of AI in microgrids are reviewed together with their benefits, drawbacks, and prospects for the future. The coordination and maximum penetration of renewable energy, solar PV, and wind in a hybrid microgrid under fault outbreaks are furthermore reviewed.

1. Introduction

Microgrids (MGs) are advancing in terms of intelligence, distribution, and flexibility. Electrical grids are being dominated by cutting-edge power electronics and artificial intelligence (AI) techniques, and this trend may continue for many years to come [1]. The increasing application of cutting-edge AI approaches in MG controls is significant and provides smooth distributed generation (DG) integration and disconnection, improved power quality for end users, and greater transient stability of the power system. As a result, MGs are becoming a trusted and more environmentally friendly source of energy. By emphasizing the independence of local energy supply, MGs that are mostly built using renewable energy sources (RESs) and distributed energy resources (DERs), commonly referred to as DGs, can combine RESs such as wind turbines and photovoltaics (PVs) with traditional energy sources such as combustion gas turbines and diesel engines. An MG is the combination of groupings of DERs and energy storage systems (ESSs), such as batteries, capacitors, hydrogen, and flywheels, with loads that are connected via a regional electric power grid [2,3]. Systems that combine two or more RESs with traditional energy sources are known as hybrid renewable energy systems (HRESs) and are frequently abbreviated as MGs [4,5].

Digitalization and automation have had a significant impact on various industries recently because of advancements in information and communication technologies (ICT). The key forces behind this transformation include significant advancements in the Internet of Things (IoT), artificial intelligence (AI), cyberphysical systems (CPSs), and big data analytics (BDA) [6,7,8]. The next-generation systems, as described by the Industry 4.0 paradigm, are the result of the evolution and convergence of new technologies such as onboard computation, intelligent and fast controllers, big data analytics, machine learning (ML), and IoT technologies [9]. As a result of these developments, it is now possible to continuously collect, process, and analyze real-time data streams in conjunction with high-fidelity models to produce a digital representation of a complex system that offers an abundance of information about its present and potential future operating conditions. The development of a new technology field known as the “Internet of Things” (IoT) has been assisted by technological advancements in hardware, software, and wireless communication. The word “things” in “IoT” are connected “smart” objects, such as sensors, machines, and vehicles, with the ability to process information intelligently [1,4]. Smart homes, smart cities, intelligent transportation, intelligent energy production, and intelligent agriculture are a few intriguing IoT applications [9,10]. IoT technology is essential to modern industry since it automates processes and lowers costs. Figure 1 shows the overview transmission of power from smart grid comprising a grid and PV-wind supply to the customers using AI control strategies.

In recent years, the field of computer science known as artificial intelligence (AI) has gained popularity. In the context of microgrids, AI has important applications that can effectively utilize the data that are available and aid in decision making in challenging real-world situations for a safer and more dependable control and operation of the microgrids. The single- to multi-microgrid network-regulating environment can be taken advantage of thanks to improvements in AI-based algorithms and computational capability with a significant number of data-processing capabilities. Important AI subcategories include machine learning (ML) and deep learning (DL). According to the input training data, ML and DL models can generally be either supervised or unsupervised. To more effectively handle the observability and controllability challenges in the context of microgrids, the system control and analysis requires a sophisticated methodology that not only relies on the physical model but also incorporates data-driven modeling [11]. The microgrid hierarchical control system has many control levels depending on the functionality to be addressed while taking into account the amount of control, communication needs and energy resources. A microgrid environment can be controlled and operated with more accuracy, speed, and efficiency by combining AI approaches with these existing schemes [12].

Building an affordable, reliable, and ecologically friendly energy system is becoming more and more important. The inclusion of distributed power generation based on renewable energy sources is part of the solution required to reach the aforementioned goal. If certain characteristics and requirements are met, well-defined areas of distributed energy generation can be regarded as microgrids (MGs). However, there are still a number of problems that need to be resolved before a large number of MGs powered by renewable energy sources can be integrated into the current electrical network. Numerous studies offer various solutions for dealing with the various control issues that MGs raise while ensuring the quality of the energy produced. In [13], an overview of the well-known droop control method used on MGs, including virtual impedance loop-based droop control, is provided. In addition to this, numerous studies [14,15,16] have concentrated on a variety of control techniques for various modes of operation of MGs. However, the energy system becomes more uncertain and complex as a result of the growing use of renewable energy sources and the active involvement of consumers. Due to the difficulties in resolving uncertainty and partial observability issues, system analysis and control cannot rely exclusively on physical modeling and numerical calculations.

Artificial neural networks (ANN), fuzzy logic (FL), genetic algorithms (GA), particle swarm optimization (PSO), metaheuristic methods, grey wolf optimization, ant colony optimization, neurofuzzy inference systems, and evolutionary algorithms are a few of the artificial intelligence-based control strategies gaining more attention [17]. Since they can adjust to uncertainties and be utilized in situations when the precise model of a system is unavailable or prone to changes, intelligent controllers are especially well-suited for these kinds of applications. A variety of studies are being conducted in this area, and market recommendations are constantly changing. By producing green energy options for the reduction of hazardous gas emissions and managing the peak load graph, the application of renewable energy sources (RES) can offer an alternative source to the reliance on fossil fuels in compliance with the international target for the environment. Future power systems can support the incorporation of RES with the use of smart grid (SG) technologies. With a high penetration of distributed generation in power systems and advancements in information and communication technology (ICT) associated with customer data, the electric power grid can be transformed [18]. Artificial intelligence (AI)-enabled smart energy markets can make it simpler to establish effective policy incentives and enable consumers and utilities to make decisions about their consumption and generation in a way that reduces CO₂ emissions. Some of the challenges facing AI in the electrical power system include designing automation technologies for heterogeneous devices that learn to adjust their consumption against pricing signals with user constraints, creating channels of communication between humans and controllers, and creating simulation and prediction tools for customers and suppliers. Intelligent tools and methods are required to manage the system properly and make choices on time as the energy sector becomes more complicated.

To handle issues of categorization, forecasting, networking, optimization, and control strategies, well-known AI approaches include the artificial neural network (ANN), reinforcement learning (RL), genetic algorithms (GA), and multiagent systems [19]. Because there are not enough sophisticated automatic controlled resources, many system operations are still carried out manually or with only the most basic automation. However, the introduction of AI into the grid system would bring about breakthroughs and provide the electrical grid new directions. Figure 1 displays the whole distributed SG concept with AI methods. Costs are reduced by optimizing controlled loads using intelligent ways. To maximize the controllable loads, for instance, Neves et al. [20] offer a genetic algorithm for the management of standalone microgrids (MGs). AI techniques provide considerably more effective and powerful ways to deal with the limitations of the conventional grid system as a result of advancements in computer power and readily available data storage. Additionally, the use of distributed computing technologies in Singapore has given rise to numerous security concerns. Threats such as physical and cyberattacks can result in infrastructure failure, privacy breaches, disruptions, and denial of service (DoS) [21]. This paper examines the smart grid’s current developments and difficulties, distributed intelligence for energy production in the future, and the function of distributed energy resources (DERs) in that system.

Different quick and intelligent fault detections and classifications, including localization and fault direction identification, protection, and coordination methods, have been explored and proposed by many researchers to address the aforementioned problems and difficulties. Determining the source of the defect and isolating it are the three key components of fault diagnostics [22]. The use of a positive sequence component [23] and zero-sequence component [24] in fault detection with protection strategies based on sequence components has been proposed. The MG protection system, according to Ustun et al. in [25], is based on a communication system and a centralized coordinated protection unit. The adaptive and differential protection approach was proposed in [26]. However, the fundamental issue with such techniques is their high cost and communication network failure.

Worth noting in this review paper is the aim of this paper, which is to provide a concise review of current AI techniques utilized for optimization and power quality management in microgrids with energy storage systems for both standalone and grid connected systems. Numerous scholars in the past focused on the optimization strategies under normal conditions; therefore, the main contribution of this paper is a review of the applications of different AI optimization strategies under fault conditions in microgrids. This paper highlights more applications of AI, and similarly, the behavior under transient and steady-state conditions.

This review article contains:

Section 2, which presents a comprehensive review on microgrid control and describes the microgrid scope, operational features, and control scheme methodologies;

Section 3 contains the summary of artificial intelligence (AI) strategies, and an overview of optimization techniques is presented;

Section 4 provides reviewed applications of artificial intelligence (AI) strategies in microgrids for optimization and a summary of various studies presenting the implementation of AI strategies;

Section 5 evaluates the Review of Applications of Artificial Intelligence (AI) Strategies under faults outbreaks in microgrids and summarizes various studies’ findings on ANN, PSO, FL, GA, and ABC techniques;

Section 6 is the conclusion of this paper, and projects future prospects and possible improvements.

2. Comprehensive Review on Microgrid Control

The microgrid can provide improved power supply quality, greater energy efficiency, and better service reliability when enhanced with contemporary power electronics-based technology [27,28,29]. The traditional way of controlling the flow of power is eliminated by the integration of multiple microsources and a new framework. Because it covers a broad range of power electronics and power system topics, research on microgrids has become increasingly popular [30,31]. The idea of a microgrid is put out primarily for two reasons: to regulate the impact of distributed generators and to make traditional grids more compatible with widespread deployment of distributed generators. Microgrids help to maintain service quality while also ensuring continuous service. The microgrid is capable of low-voltage operation and has a variety of distributed energy resources. It is capable of functioning in grid mode or off grid mode [32]. The implementation of different smart grid functions, such as digital and two-way communication, distributed generation, self-monitoring, self-healing, adaptive and islanding mode, remote check and intuitive control, can be sped up with the help of microgrids [33].

The microgrid may successfully address a number of power generation-related problems and is a complementary suppression for constrained fossil fuels [34]. Operation in regular and island modes, plug-and-play functionality, protection, power quality, security, voltage and frequency control, system stability, and energy management are a few of the most important of these problems. Despite their numerous advantages, microgrids present numerous technological obstacles, and protection is one that needs particular focus. For a microgrid to operate consistently and profitably, it must be protected. The protection plan must be capable of handling any fault without interfering with the overall structure. It should run in the shortest time possible. It must fulfill the conditions for both the grid-tied and the islanded modes. The steps in the protection scheme process involve locating the problem, cutting off the problematic area from the rest of the system, and fixing it as quickly as possible. Therefore, the design of the protection system must be precise [35,36,37]. A microgrid is a self-controlled individual entity that has a “plug-and-play” capability for each distributed energy supply that satisfies the demands of local loads. As a result of all of these qualities, MGs can evolve into complex systems with variable degrees of uncertainty and the requirement for complicated controllers. Frequency regulation, voltage regulation, load sharing and tracking, and fault protection are some of the primary controls that must be considered for an MG.

2.1. Microgrid Control

2.1.1. Conventional Control

The primary operational and control characteristics of the traditional control architecture are reviewed. From the comparisons, it can be concluded that distributed control approaches will be crucial in decarbonizing the future distribution or island grid as DER penetration in the distribution network increases. Although it is very complicated to accomplish, it is very successful in terms of control features.

2.1.2. Unconventional Control

A virtual impedance control loop can be added to the primary layer to simulate the actual behavior of the system, which is often responsible for droop control to stabilize and dampen the system. Local controllers carry out the intricate controlling; therefore, this layer can respond quickly or in real time. The use of a main control, in which one convertor serves as the master and the others as slaves, is also suggested in [38].

2.2. Microgrid Mode of Operation

Microgrids can run in one of two ways: connected to the grid or as an island. Each operating mode has a separate set of operational needs. The following are the modes.

2.2.1. Grid-Connected Mode

Without any disruptions to the main grid’s power quality leading to power outages, it is otherwise in the regular working mode. In this mode, the microgrid may either import or export power to the main grid and deliver power to its full local load, depending on the total amount of power generation of the local load. The microgrid also keeps the electricity flow in this mode bidirectional.

2.2.2. Standalone Mode

The microgrid disconnects from the main grid and runs independently whenever there is a failure or change in power quality on the main grid. The microgrid will sustain high-quality power and could give users a continuous supply if there is a problem with the main grid causing power outages. Furthermore, the microgrid can easily be islanded from the main grid if other disturbances such as frequency drops, voltage sags, or any fault develop in the main grid [39,40,41].

2.3. Microgrid Frameworks

The microgrid may either import or export power to the main grid and deliver power to its full local load, depending on the total amount required for users. Because microgrids also manage and distribute the flow of electricity to users, they can be thought of as a scaled-down version of the existing centralized electrical system. However, it is carried out locally, unlike the normal approach. It is possible to think of it as a single aggregated load in a power system because it only has one regulated unit [42]. A microgrid is an emerging idea that describes a small power system with a group of dispersed generators working together with correct energy management, protection devices, control devices, loads, and related software [43,44]. The two most important characteristics of microgrids are:

• Peer-to-peer, which means that the operation of the microgrid is not dependent on the availability of a specific component, such as a master controller or a central storage system.
• Plug-and-play, which allows DG sources to be placed anywhere in the microgrid without having to change the protection scheme. This functionality makes it easier to install developing DG sources and lowers the chance of microgrid engineering failures.

Hybrid microgrids are made up of the individual DC and AC microgrid architectures. Consequently, hybrid AC-DC microgrid contains both the AC and DC microgrid’s advantages. Figure 2 displays a genuine hybrid DC-AC microgrid architecture. Connecting AC and DC microgrids makes use of bidirectional AC-DC converters. For linking DC power generators, connecting PV panels, wind energy systems, and energy storage systems (ESS) are used, and they connect to the battery in this case, and there are loads connected from the system. For greater efficiency, photovoltaic (PV) panels connect to the DC microgrid. DC-DC boost converters are used when connecting this system for simulation of greater stability performances.

2.4. Importance of Microgrids

(i)

Microgrids enable distributed generation and high penetration of renewable energy sources.

(ii)

Microgrids support adequate generation since they can manage internal loads and generation.

(iii)

Microgrids strengthen local economies, and their structures will attract small businesses and provide additional jobs in the area [40,45].

(iv)

Areas with microgrids will continue to receive regular power supplies during natural disasters, outages, etc.

(v)

If a microgrid can meet local demand, transmission and distribution losses in the power system are less expensive, and the cost of expanding transmission and distribution is also lowered.

(vi)

Because microgrids make use of ecofriendly renewable power generation techniques, they will aid in lowering CO₂ emissions.

(vii)

It provides power to the main grid when the microgrid produces excess energy;

(viii)

Stability is altered by the microgrid.

(ix)

Compared to traditional power generation, the cost of energy produced by microgrids with distributed generating assistance is lower [46].

2.5. AC Microgrid Overview

In general, mixed loads (DC and AC loads), distributed generation, and energy storage devices are coupled to the common AC bus used in AC microgrids. Since most loads and the grid itself are AC, AC microgrids can be simply integrated into a traditional AC grid. As a result, it is more flexible, capable, and controllable. However, when a DC/AC converter is used to connect DC loads, DC sources and energy storage devices to the AC bus, the efficiency is intensely reduced [35,47,48]. The primary elements that need to be synchronized in AC microgrids are active power, reactive power, imbalance components, and harmonics. DC power is the primary element that needs to be controlled in DC microgrids. Consequently, the DC microgrid control system is easier to use than the AC microgrid control system [49]. In AC microgrid design before the connections, photovoltaic (PV) system-generated DC power is converted into AC using DC-AC inverters. Rectifiers convert AC power into DC power so that it can be utilized to power DC loads. Without any transformation, the AC load receives a direct supply from the AC bus. Converters use the connection between the AC bus and the wind power generation system to manage both active and reactive power. Because just phase matching is needed between the main grid and the AC microgrid, connecting to the main grid becomes simpler.

2.5.1. Advantages of AC Microgrid

(a)

The use of high-efficiency transformers. For distribution purposes and for the nearby local loads, the voltage of AC microgrids can be increased and decreased using transformers, respectively.

(b)

Protection techniques for AC circuits are favorable due to periodic zero voltage crossings since switching circuit breakers extinguish the fault current arc at zero crossings.

(c)

Stable voltage can be achieved by independently managing reactive power.

(d)

In grid-tied mode, the AC microgrid will automatically disconnect if any fault conditions arise in the microgrid. Since the AC load receives a direct supply from the AC microgrid, any disturbances in the main grid will not affect it [50].

2.5.2. Disadvantages of AC Microgrid

(a)

To power DC loads such as battery charging, computers, DC fluorescent lights, etc., AC power must be converted into DC power. These conversions result in a decrease in efficiency.

(b)

The use of power electronic converters causes an introduction of harmonics introduced into the main grid.

(c)

The DC output of renewable energy sources must be converted to AC using inverters, which makes interconnection difficult [51,52,53].

2.6. DC Microgrid Overview

A common DC bus connected to the grid via an AC/DC converter is utilized in a DC microgrid. AC microgrids and DC microgrids both operate under similar principles. Since a DC microgrid only requires one power conversion to link a DC bus, it is a good alternative to an AC microgrid for reducing power conversion losses. As a result, the system efficiency, cost, and system size of a DC microgrid are higher. As a result of the insufficiency of reactive power, DC microgrids are also more stable and better suited for integrating distributed energy resources (DERs) [47]. In order to reduce the cost of the power electronics converter and boost efficiency, diverse DC loads can be directly linked with the DC microgrid bus without any transformation. DC-AC converters are necessary for connecting AC loads. Research on DC microgrids is gaining traction due to the growth of DC renewable energy sources.

2.6.1. Advantages of a DC Microgrid

(a)

The direct connection of a battery storage system to a power source for backup is possible. In times of peak load or in the absence of any distributed generators, a backup storage system will provide power.

(b)

Direct connecting lowers the need for several power conversions and boosts system effectiveness.

(c)

It enables easy connection of renewable energy sources.

(d)

Should there be a power outage in the AC main grid, the DC microgrid’s battery storage will routinely supply electricity to loads.

(e)

The running costs and power converter loss of a DC system can be kept to a minimum because all that is needed to connect to the AC main grid is a straightforward inverter unit.

2.6.2. Disadvantages of a DC Microgrid

(a)

Most load units in the current power system configuration demand AC power. Therefore, a DC-only distribution network is not practical.

(b)

Compared to an AC system, voltage transformation in a DC system is less systematic.

(c)

The integration of AC generators necessitates the use of a rectifier to convert AC power to DC power [54,55].

2.7. Comparison of AC and DC Microgrid Conversion

The difference between AC loads and DC load conversion steps is under investigation by numerous scholars who are interested in the design of hybrid renewable energy systems and comparing the benefits of deploying the best form. Therefore, there are different conversion steps, and these come with different pros and cons.

Table 1. Comparison of AC and DC microgrid conversion steps.

Type of Microgrid	DC Load	AC Load
DC microgrid	Single conversion	Multiple conversions
AC microgrid	Multiple conversions	Single conversion

presents the comparison of AC and DC microgrid loads.

2.8. Hybrid Microgrid

There are numerous scholars who are interested in the design of hybrid renewable energy systems. Consequently, there is a ton of material on this subject that can be used. The aforementioned design subject relates to energy systems, where it is observed that the best distribution and optimum placement, kind, and size of generation components have been established on particular nodes. As a result, this kind of system can load the requirements at the lowest possible cost [56]. The idea of hybrid renewable energy can calculate the price and output over the technology’s lifetime. The initial estimate for the lifetime cost often comprises two components: the operational cost, such as the primary cost; and the preservation cost, both of which amount to a “fixed cost”. Furthermore, when calculating the lifespan cost, the financial values are updated according to the timing, and should be considered. As such, the optimal hybrid system designs mix producer kinds and sizes to achieve the lowest possible lifespan cost and productivity. Thus, the “optimal configuration” or “optimal design” is defined as the design with the lowest “net present value” (or NPV), with all possible hybrid system designs being in optimal transition [57,58]. To act as real-time system integration, there are numerous ways to provide an “optimal design” indicator and numerous commercially available software products. In addition, many optimal strategies are used by numerous researchers for hybrid renewable energy system sizing. In order to optimize hybrid PV/wind energy systems, researchers have used a variety of optimization techniques, including graphical construction [59,60], probabilistic techniques [61], iterative approaches, dynamic programming, artificial intelligence (AI), linear programming, and multiobjective optimization [62].

DC-DC buck converters are used to link DC loads to the DC microgrid, such as fluorescent lighting and electric automobiles. Energy storage devices are connected to the DC microgrid using bidirectional DC-DC converters [63]. The AC microgrid is also directly connected to AC loads, such as AC motors. When an AC microgrid is overburdened, power will switch over to a DC microgrid [64]. The main converter will serve as an inverter during this operation. Power will flow from the AC microgrid to the DC microgrid during the overloaded state of the DC microgrid, and the interlinking converter will serve as a rectifier [65]. Bidirectional AC/DC converters’ primary goal is to regulate power flow between DC and AC microgrids while stabilizing their respective DC and AC bus voltages and frequencies [66,67,68].

Hybrid model techniques are a beneficial collaboration of two or more separate methods that employ the beneficial effects of the methods to achieve the best possible result for a specific design problem. Because many of the challenges we face are multiobjective in nature, implementing a hybrid technique is an ideal aim in nature, and adopting a hybrid approach is a suitable alternative method to address problems that necessitate a thorough understanding of all the methods. To handle a multiobjective problem that encompasses costs, environmental consequences, fuel price risks, and imported fuel, Meza et al. [69] established a multiobjective model to generate expansion planning (MGEP) and an analytical hierarchy process (AHP) model.

2.9. Energy Storage System

Electrochemical systems (batteries and flow batteries), kinetic energy storage systems (flywheels), and potential energy storage are the three categories into which energy storage devices (ESS) can be divided (pumped hydro and compressed air storage). A thorough comparison of various energy storage technologies may be found in References [70,71,72,73]. Small-scale renewable energy systems cannot use pumped hydro storage or compressed air energy storage systems since these large-scale energy storage devices are typically used in high power ranges for regular power systems. In microgrid applications, energy storage devices may enhance the power quality, reliability, and stability between loads and the output of distributed generated resources. According to the characteristics of loads and dispersed energy resources, it is possible to identify more suitable energy storage devices.

The following is a summary of some important energy storage technologies suitable for MG applications: One of the most popular forms of energy storage is batteries. They fall under the categories of lithium-ion (Li-on), lead acid, nickel cadmium (Ni-Cd), and nickel metal hydride batteries. Long-term energy storage is possible using lead acid batteries, despite their poor performance and short cycle life (1200–1800 cycles). Ni-Cd batteries have higher energy densities, longer cycle lives, and require less maintenance when compared to lead acid batteries. However, its biggest drawback is a large initial capital expenditure. The energy density of NiMh batteries is around 25–30% higher than that of NiCd batteries, and they also have a cycle life that is comparable to that of lead acid batteries. In comparison to lead acid, Ni-Cd, and NiMh batteries, Li-on batteries have the best energy density; nevertheless, the investment cost and short life span are the key downsides of Li-on batteries [74,75].

To lessen the negative effects of PV integration, a battery storage system integrated with solar PV systems has been presented in Reference [76]. Simulink and Homer analyses were conducted in Ref. [74] to evaluate various battery storage solutions from a technoeconomic standpoint. Flywheel energy storage systems have a long lifespan, a high energy density, and a high power density. However, the disadvantage of flywheel energy storage is that substantial friction losses are likely to occur. They can be applied to lessen the erratic power output of solar and wind power systems. In Refs. [77,78], flywheel storage systems and a diesel generator are utilized to provide a UPS service to the critical loads. Supercapacitors, which are based on the characteristics of capacitors and electrochemical batteries without a chemical process, are also known as ultracapacitors or electric double-layer capacitors.

The use of porous membranes, which enables ion movement between two electrodes and offers direct energy storage while also reducing response time, is the primary distinction between a capacitor and a supercapacitor [74]. Additionally, its energy density and capacitance values may be hundreds to thousands of times greater than those of the capacitors. Supercapacitors have a longer cycle life, better power density, and an energy efficiency of roughly 75–80% when compared to lead acid batteries, which have a lower energy density. However, the biggest drawback of this technology is its expensive price, which is almost five times greater than that of lead acid batteries [75]. Supercapacitors have been presented as a good option for reducing the natural variations prevalent in the intermittent renewable energy sources of wind and wave, respectively. SMES systems feature extremely long lifespans (tens to thousands of cycles), extremely high efficiency (up to 95%), quick response times, and expensive implementation costs. Power factor optimization, frequency management, transient stability, and power quality enhancement are potential applications [79,80].

2.10. Faults and Protection in Microgrid

A hybrid microgrid’s faults can be classified as load, converter, generator side, sensor and cyberattack-initiated problem [81]. Pole–pole faults and pole–ground faults all occur at the DC load side in a hybrid microgrid [82]. The AC side experiences conventional faults, such as line to ground (L-G), line to line (L-L), line to line to ground (L-L-G), and line to line to line (L-L-L), while the AC microgrid may also experience converter faults. Given that the dispersed generator is located close to the load center, the likelihood of DC transmission line faults is lower than the likelihood of AC line failures. By using standard detection and isolation procedures, the AC side problems can be found. When a problem occurs on the DER or load side of a DC microgrid, it is important to determine if the system is in islanded mode or grid-connected mode initially. As RES cannot supply very high fault currents where synchronous generators typically deliver, the threshold value of a fault current is lower when the system is operating in standalone mode. The threshold signal value should be revised in this situation.

In a DC microgrid, converter faults are caused by either an open circuit or a short circuit in the power converter switches. Since heavy current does not flow through an open circuit when it occurs, protective devices cannot detect this problem using the overcurrent phenomena, making open circuit detection more difficult than short circuit detection. In a level 3 boost connected grid [83], a unique control approach is employed to resolve this issue. According to this method, the boost converter functions as a 3 level boost converter under normal circumstances while acting as a 2 level boost converter in abnormal circumstances [84]. The system is resilient enough to deliver half of the load because the supply is unbroken even when it is malfunctioning. In an active hybrid system, the various defects include sensor faults and problems brought on by cyberattacks. Actual values of the system parameters are provided by sensors for various controller actions. In the event that the system’s characteristics are not accurately represented by the sensor output, the controller will process misleading data and send undesirable pulse signals to the converters, which will cause the system to behave incorrectly and lose stability. The grid side converter fails to maintain current and voltage in such a circumstance. The adversary can launch a cyberattack by overwriting the sensor value in order to impose the broken state in the grid.

The sliding mode oberver technique is used to estimate error due to sensor failure or a defect caused by a cyberattack in a hybrid microgrid attack [81]. The absence of zero crossing current, reliance on converter topologies, output filter effect, and system grounding provide obstacles to DC microgrid protection, which must be addressed in a new protection scheme. A big amount of arc forms once a protection device quickly breaks a high current. Because AC has a natural zero crossing of the current, fault isolation is performed at the zero current occurrence, which results in the lowest amount of arc production. In contrast to an AC system, a DC system lacks a natural zero crossing, hence the arc’s strength is much higher. Despite several benefits, DC microgrid protection faces several difficulties. Some common issues to AC and DC systems include the inability of overcurrent relays to protect microgrids or limited fault currents in islanded mode. Despite this, there are certain additional problems that impair microgrid protection.

3. Artificial Intelligence (AI) Strategies

John McCarthy, a computer scientist, created the term “artificial intelligence” [85] in 1979 and later defined it as “the science and engineering of constructing intelligent machines” [86]. The definition of artificial intelligence is teaching computers to resemble human thought processes and even emulate human behavior [87]. A data-driven system that enables a computer or software to carry out tasks or make judgments is also an area of computer science devoted to replicating human cognitive processes. In their book “The Race Against the Machine” [87], Brynjolfsson and McAfee argued that due to major advancements in AI and ML, human civilization will not be capable of supporting anything like full employment. The routing algorithm is one of the most important characteristics of computer networks because it is responsible for delivering data packets from source to destination nodes. Although static and centralized routing algorithms could satisfy industrial-age networks, they are incapable of responding to new-age networks that deal with massive amounts of data and a large number of heterogeneous users and traveling entities. AI has numerous applications in power system engineering. Artificial intelligence (AI) and advanced numerical computation software are critical. Sensors have a wide range of applications in industrial settings, including safety enhancement, data acquisition, and environmental and human body monitoring. Wireless sensor networks (WSNs) may use multiple antennas to transmit sensing data. A previously unexplored path for the shrinking of wireless sensors has now been made possible by additional advancements in the design of planar antennas. Designers may create networks of interconnected sensors and cameras using AI technology, ensuring that an autonomous vehicle’s AI can “see” its surroundings. They must also make sure that information from these on-board sensors is transmitted quickly. Utilizing the capabilities of AI applications may also make it possible to improve system performance while solving issues more effectively. They have the chance to adjust how their organizations run day-to-day business and develop over time.

3.1. Advantages of Artificial Intelligence

Superior demand forecasting, direct automation, a safer workplace, sophisticated root cause analysis, smooth 24/7 production, a smarter workforce, increased workplace safety, lower operational costs, improved production defect identification, automated quality management, higher operational efficiencies, better power product or equipment design, quick decision making, predictive maintenance, testing, and qualification are just a few of the benefits of artificial intelligence [88].

3.2. Strategies of Artificial Intelligence to Support the Optimization of Hybrid Energy Systems

3.2.1. Genetic Algorithm

Utilizing fitness (objective) function and constraints, the genetic algorithm (GA) generates the best solution from the available solution domain [89,90]. GA can be divided into two categories: single-objective function optimization and multiobjective function optimization. Single-objective function optimization produces a unique solution that can be either maximum or minimal by optimizing just one objective function. However, in the optimization of a multiobjective function, many objective functions are optimized, leading to a solution space that is not dominated (from the entire feasible solution domain, a sub domain is obtained, which is better than the remaining solution space). GA encompasses the three fundamental genetic processes of crossing, mutation, and selection.

Choosing the population size is the initial stage in using GA to find a feasible solution. The population size can be chosen at random [91] or based on our knowledge of a possible solution. The value of the objective function is examined at each population size point. The population size’s points that produce an accurate fitness value are regarded as parents in GA. The new population known as children is created using these parents [92]. Crossover and mutation are two methods that could be used to accomplish this [93,94]. While mutation creates various changes to the present parents based on fitness value, crossover is the combination of many parents. In many instances, elite offspring with the same parents may be produced (in those cases when the selected initial population gives an accurate fitness value). Up until a workable solution is found that provides an accurate fitness value, GA creates new populations. The algorithm is stopped using specified stopping criteria after the accurate fitness value is obtained. Time limit, fitness limit, generation size, function toleration, and constraint tolerance are just a few examples of stopping criteria for GA. Numerous published studies discuss the use of GA in hybrid systems for sizing research. For instance, Yang et al. [95,96] and Bilal et al. [97] used GAs to size a PV hybrid wind system. The deployment of hybrid PV/wind systems can result in reduced system expenses when compared to systems in which the sources of only PV and WG are applied, according to Koutroulis et al. [98,99], who employed genetic algorithms to determine the energy costs of the entire system. On the other hand, Lagorse et al. [100] used the concept of a multisource hybrid source combining wind and oil to make use of GAs. The authors of [101] reported that their use of genetic algorithms resulted in improved battery size. Additionally, Lopez et al. in [102,103,104] created a simulation application called Hybrid Optimizations based on Genetic Algorithms (HOGA) to program compounds of standalone hybrid energy systems with content from renewable sources and conventional diesel generators. Lagorse et al. employed a hybrid GA and simplex-based approach [105], whereas Zhao et al. [106] designed a GA that has, as its primary components, a wind source and essential technical specifications. These served as the parameters for the input. To lower production costs and increase the dependability of the system, a wind source of energy was also constructed. Furthermore, Li et al. [106] made use of a GA to increase the rate of gearbox proportioning, which in turn increased the power ratings of the multihybrid stable wind generators. The locations of the wind turbines on the property are included in the wind outline, which aims to maximize energy production [107]. Grady et al. [108] provided a GA to determine the best generation of wind turbines with the highest output capacity while limiting the number of fixed turbines and the amount of land that each wind farm occupied. The solution to this difficulty was developed by Emami and Noghreh [109] using a fresh coding strategy and a brand-new objective function with GAs.

For the management of the cost, electricity, and efficacy of the wind farm, their solution outperformed earlier methods that were suggested. In order to tackle both active control algorithms for the wind and optimal design challenges, Li et al. [110] employed a multilevel GA. Kalogirou [111] also employed ANNs and GAs to achieve his financial objectives in order to expand the solar energy infrastructure. The ANN method is appropriate for establishing a connection between a high container size and a collector region while using minimum power, which is ideal for the system. Then, using a GA, the optimum magnitude of factor to use to extend the life-cycle reserves is determined. By reflecting a variety of systems and operating conditions, Sharma et al. [112] used the GA approach to improve the thermal performance of flat laminate solar air warmers. In order to determine the electrical factors, Zagrouba et al. [113] used the GA technique. For identifying the equivalent highest power point, this study uses photovoltaic solar cells and modules.

3.2.2. Particle Swarm Optimization

A population-based stochastic optimization technique is called particle swarm optimization (PSO). This tactic has been influenced by practices such as fish tutoring or bird flocking. The authors of [114] carry out an ongoing or continuous EMS PSO inside an MG. The replica has been updated every three minutes. Reducing or limiting the system’s overall power expense is the goal. To determine the ideal positioning and size of the MG components to ensure the system’s dependability and profitability, the multiobjective particle swarm optimization (MOPSO) was implemented in [115]. For the system configuration and dimensioning, a multiobjective PSO methodology is presented [116]. Reliability (unwavering quality), running costs, and natural effects are three competing criteria that must all be balanced for an MG to operate at its best.

Refocusing optimal energy management for industrial MG was put forth by Li et al. [117] and is based on PSO. It may be used in standalone and network-associated modes. Keeping MGs operating and maintenance costs to a minimum is the goal of the standalone mode. However, in network-associated mode, it also increases the benefit of power exchanging with the main network. In terms of the overall optimal solution and computation time, the suggested approach produces better results when compared to the genetic algorithm. PSO has recently solved a significant problem in the power consumption of high-gain antennas. This nature-based optimization algorithm (i.e., PSO) also has reduced power dissipation in long-distance communications, as discussed in “Multi-objective particle swarm optimization to design a time-delay equalizer met surface for an electromagnetic band-gap resonator antenna”. These aspects show how PSO is making changes in the electrical engineering field.

3.2.3. Artificial Neural Network

An artificial neural network (ANN) is a replication of a biological neural network that efficiently links different parameters to a greater number of uncertain data points. To relate different parameters, ANN models do not need complex mathematical bases or equations [118]. As a result, ANN takes less computational work when coupling any number of parameters with a large number of uncertain data points than previous approaches [119,120,121]. The process of training an ANN using imported data is known as supervised learning, or training. Like the neurons in the human brain, ANN has a number of these [122,123,124]. Weight is a fractional value that connects these neurons to one another [122,125]. The values of weight become constant when error reaches a reasonable level during the training phase in order to predict correct results [126].

The basic framework of the ANN model is depicted in Figure 3. The input layer, hidden layer, and output layer of an ANN are each made up of neurons [127,128,129]. The quantity of input parameters determines how many neurons are in the input layer, and the quantity of output parameters determines how many neurons are in the output layer [130]. The number of neurons in the hidden layer is typically determined using (1), which depends on the quantity of input + output neurons and the quantity of training data points [131]. The number of hidden neurons is frequently determined by using the trial-and-error method [91]. ANN has a wide application in various industries, such as space exploration.

$$\begin{array}{l}Numberofhiddenneurons=\frac{numberofinputs+outputneurons}{2}\\ +\sqrt{numberoftrainingdatapoints}\end{array}$$

(1)

The time-dependent neural network is very important, where t stands for time and bias is a second parameter used to modify the neural network’s output. The inputs could be added to the network at different times, according to the right-side neural network. The entire collection of data for an input–output parameter is split into two groups: the training data set, which contains a larger percentage of the data points, and the validation data set, which contains the remaining data points and is used to verify the trained neural network [132,133,134]. Neural networks import input–output parameters together with their training data points. The training of this network continues until the acceptable error is reached. Once the allowable error is determined, the trained network is validated by importing the values of the validation data set’s input parameters and predicting the corresponding values of the output parameters. If the difference between the actual and predicted results is within the allowable limit, the trained neural network may be suggested as the best neural network for the prediction. These predicted values of the output parameter of the validation data set are compared with corresponding actual values of the output parameter of the validation data set [134,135]. Different training algorithms are used to train neural networks, and these algorithms include training function, learning variant, transfer function, and the number of hidden neurons [136].

There are several different training functions, learning variants, and transfer functions accessible [137]. A reasonable number of epochs are trained for the outputs. The neural network predicts the values of the output parameter for the associated input values for the appropriate training algorithm and training epochs. The trained neural network with that combination of training algorithm may be chosen as the best neural network with the best training algorithm if the error value is smaller than the allowable value. When the error is bigger, the neural network is trained either using the same training algorithm for a different number of epochs or using a different training strategy until the error is below acceptable limits [138]. The predictions made by the best neural network utilizing validation data points demonstrate that the trained neural network generalizes well in [139].

Every neural network has a threshold value and an output value is only predicted when the sum of all the input values multiplied by their associated weight values exceeds the threshold value, as shown by (2) [121,138,140].

$$Y_k={\displaystyle {\sum }_{i=1}^m{W}_{ki}xi}$$

(2)

where $W_{ki}$ is the weight at the ith neuron, and $xi$ is the value of the ith neuron. $Y_k$ is the neural network output. There are numerous types of neural networks that are available and are detailed based on various topologies and properties.

The ANN determines a related output when an input training pattern is introduced. It is compared to the output that is already available. The optimization technique makes use of this difference to train the network. The ANN determines a related output when an input training pattern is introduced. It is compared to the output that is already available. The optimization technique makes use of this difference to train the network.

3.2.4. Fuzzy Logic

Instead of dealing with data points, fuzzy logic (FL) deals with the ranges of the various parameters. FL could therefore anticipate precise outcomes for all data points within the limits of different parameters [141]. The human understanding is what determines FL accuracy [142]. The general fuzzy model diagram is displayed in Figure 4. Two modules, Mamdani and Sugeno, make up FL. Unlike the Sugeno module, where input parameters have a variety of ranges and output parameters include data points, the Mamdani module of FL divides both input and output parameters into ranges [143]. FL is divided into two portions, the first of which presents the influencing (input) parameters and the second of which displays the performance (output) parameters. Each input/output parameter’s whole range is broken down into a number of minor ranges [144]. A suitable basic shape, such as a triangular, trapezoidal, sinusoidal, or Gaussian, is used to represent each small range of variation [145]; these shapes are shown in Figure 4.

The choice of this basic form for any range is made depending on how frequently the values in that range vary. A membership function is a range with the appropriate primitive shape [146]. The number of membership functions and the border range of each membership function depend on how imported input–output data behave. Following the selection of membership functions, rules are created in the fuzzy modular’s rule editor based on the values of the input–output parameter. Data sets are divided into training and validation data sets to ensure the accuracy of fuzzy models that have been developed [120,132,133]. If-then statements and “AND,” “OR,” or “NOR” Booleans are used in rules to link the input and output values [147,148]. The connection between the input and the relevant output values is thus made by rules. Because there are fewer rules due to the smaller quantity of data points, it may anticipate erroneous results for some imported input values. Larger numbers of data points result in more rules, which improves accuracy, but occasionally creates overfitting (where there may be several output values for a single imported input value), which also forecasts erroneous results [142,148].

In the fuzzy modular’s rule viewer, rules are put into action. The rule viewer is divided into two sections, the first of which contains all the input parameters and their membership functions, and the second of which contains the output parameters and their membership functions. Output values are predicted based on the chosen rules by importing input values from the training data set. There are numerous defuzzification methods, including centroid, last of maximum, mean of maxima, middle of maximum, center of area, center of gravity, and bisector of area, among others. Additionally, the user can create their own defuzzification methods and membership functions (type, number etc.) [135].

3.2.5. Artificial Bee Colony

The metaheuristic algorithm artificial bee colony (ABC) is inspired by honey bees’ clever activities, such as collecting and distributing nectar. To avoid the local optimum and find the global optimum, ABC uses stochastic principles rather than gradient information. The ABC method was employed by the authors of [149] to address the question of the ideal microgrid size. The NPV has been optimized for the microgrid’s advantage. The methodology’s effectiveness was tested using real data on the climate and electricity demand in the east-central region of Italy to study the best evaluations for solar, wind, and energy storage systems.

The impact of AI is growing across many sectors of engineering, particularly in sustainable energy systems. AI has been used in a wide range of applications, as seen in

Table 2. Application of artificial intelligence in various power systems.

References	Applications
Zahmatkesh and Al-Turjman, 2020 [150]	IoT security and safety system in smart cities.
Pan et al., 2015 [151]	Energy-saving with the use of smart home concept.
Zhou et al., 2016 [152]	Smart energy management.
Dawn et al., 2019 [153]	Improving wind turbine and unified power flow controller (UPFC) electrical stability.
Ouahiba et al., 2018 [154]	Occupants’ satisfaction and smart buildings.
Badiei et al., 2020 [155]	New solar technologies.
Aly, 2020 [156]	Wind.
Wang et al., 2020b [157]	Solar load.
Wei et al., 2019 [158]	Gas.
Jasiński, 2020; Bedi and Toshniwal, 2019 [159,160]	Electricity price forecasting.
Erixno and Rahim, 2020; Kermadi and Berkouk, 2017 [161]	Photovoltaic generation maximal power point monitoring on a fuzzy logic basis.
Shi et al., 2020 [162]	Grid stability and reliability.
Antonopoulos et al., 2020 [163]	Demand forecast and demand side management.
Foresti et al., 2020 [164]	Predictive maintenance.
Ruhnau et al., 2020; Zhang and Yan, 2020; Wang et al., 2019 [165,166,167]	Renewable energy generation forecast.

, which presents the application of artificial intelligence in various power systems. The employed bees (EBs) and onlooker bees (OBs) make up the artificial bee population. The environment is represented by the search space in ABC, and each point in the search space corresponds to a potential food source (solution) that the artificial bees can use. The value of the function that needs to be optimized at the appropriate position indicates a food source’s quality. Initially, the EBs scout, and each EB decides to take advantage of a food source it has discovered. The number of food sources that are now being used in the system and the number of EBs are thus related.

4. Reviewed Applications of Artificial Intelligence (AI) Strategies in Microgrids for Optimization

The study of AI is a segment of computer science that advances AI software and hardware. AI has several subareas, including hybrid techniques such as ANN, FL, GA, PSO, and ABC, that combine two or more of these divisions. The appropriate usage of AI technology sets the stage for systems with strong AI performance or other qualities that might not be compatible with conventional methodologies.

4.1. Reviewed AI Optimization Strategies

Optimization methods have grown in admiration since they play a vital role to approach practical complex difficulties in electrical systems with the preferred forecast precision. The reviewed enhancement strategies and applications of artificial intelligence (AI) strategies in microgrids for optimization using solo prediction and a summary of various studies presenting the application of AI strategies is included in this study.

4.1.1. Reviewed Optimization of GA

GA is a type of global search heuristic. A specific subset of evolutionary algorithms, such as mutation, selection, and combination, use these methods to be inspired by evolutionary biology. Both of the following are necessary for a normal GA: a genetic demonstration of the problem domains, and using the fitness function (FF) to calculate the solution domains; for problem domains with the complex FF landscape and the conventional hill climbing strategy that may not succeed, GA may be helpful. For the best sizing of a standalone hybrid SPV/WTG system per year in [168], the authors used GA and elitist techniques (8760 h).

4.1.2. Reviewed Optimization of FL

The fact that the FL-developed models map the precise situation to the best level accounts for their widespread use; therefore, it is considered in many cases. This approach is used when the experts’ response is hazy. To reach consensus, professionals debate for a long time. In fuzzy regression, the dependent and independent variables’ data are both represented as fuzzy data, and the resulting regression equation calculates the effects of the independent factors on the dependent variable. The regression approach is comparable to fuzzy gray prediction in that the old area in the variables considered for dependence prediction is presented using fuzziness. To determine the variables’ relative status, fuzzy AHP and ANP are used. When ranking the variables, the fuzzy technique helps to accurately capture the fuzziness in people’s attentions. The range of the variables is given to the variables to help clearly define the clusters and establish borders. These strategies are applied to a problem domain. Fuzzy regression and fuzzy gray prediction are used for the forecasting. For determining the relative importance of energy resources, fuzzy AHP and ANP are used. Fuzzy clustering is used to gather resources that depend on factors such as cost, availability, pollution, etc. These methods, which are used to anticipate or assess the importance of electrical utility, are characterized as “simple” methods due to the complexity of the method itself.

4.1.3. Reviewed Optimization of ANN

A structured collection of AI neurons known as the optimization of ANN employs mathematical models for information processing that rely on the connectionist approach to computation. The authors of Reference [87] used ANN-based techniques to implement preemptive control strategies for large HRES. Essential components of the dynamic safety pattern class that calculate the level of safety and are superior to conventional statistical approaches are artificial neural networks (ANNs). By analyzing the peculiarity of SPV, WTG, the reference [169] provided an ANN control technique for a multienergy common DC bus hybrid power supply.

In the operational circumstances of the grids under changing load, voltage and frequency variations, as well as active–reactive (PQ) power control, are significant challenges. When renewable resources such as solar, wind, and battery storage units are operated with grid interaction and integrated into the grid, these fluctuations and power flow become more challenging to manage. Controlling the systems during the grid connection of the RES that works with the grid is challenging because of unbalanced load circumstances and abrupt power changes of the sources. Numerous issues have been discovered on both the load and source sides when issues of grid-interactive systems using RES are examined [170]. The following is a list of some of these issues:

• Voltage and frequency oscillations brought on by RES’s unreliable power generation.
• Disturbances in voltage and frequency brought on by load imbalance in systems that interface with the grid.
• PQ power transmission imbalances caused by the sources’ fluctuating load circumstances.
• The grid’s reflection of output surplus as reactive power, which happens when renewable energy systems are improperly planned.
• Connection issues that occur when RES are connected to the grid.

4.2. Reviewed Hybrid Microgrid Optimization Using Artificial Intelligence Strategies

Combining two prediction models has grown in popularity as a way to address complicated real-world problems with the desired prediction accuracy. A hybrid model is one that combines two prediction models. Each solo prediction model has its own set of restrictions. In order to produce better outcomes, hybrid models are therefore chosen to eliminate or lessen the limits of one model by fusing it with another model. Many hybrid models have been developed recently; however, according to the researched literature, ANN + FL, ANN + GA, and FL + GA are the most often used hybrid models for problems relating to solar energy. The hybrid models are explained and reviewed in this paper.

4.2.1. Applications of Fuzzy and Ann in Voltage and Frequency Fluctuations

Different techniques have been used to reduce voltage and frequency variations of grid-interactive systems working concurrently with RES employing FL and ANN algorithms. Following are some recent studies that have been published in the literature, with an emphasis on FL and ANN. An FLC is suggested by Rajesh et al. to regulate the WT’s rotor frequency in hybrid renewable energy systems. This model’s efficiency has been demonstrated by performance evaluations on a variety of controllers based on switching time, settling time, rising time, and percentage overshoots. Fathy et al. propose an adaptive neuro fuzzy inference system (ANFIS) algorithm trained through the ant lion optimizer (ALO) in order to achieve optimal load frequency control (LFC) in a system consisting of RES. In the designed system, the error signal obtained from the fuzzy system and the PI output are given as a feed input to the ANFIS controller for training with a loop. ALO-ANFIS’s results are compared with other approaches.

Data such as performance characteristics, settling time, frequency, and over and under drawn from power line deviations were calculated and compared with other reference algorithms [171]. An approach based on a dynamic dead band and frequency filter is proposed by Aziz et al. With a variable speed WT generator running in grid contact, this method offers power and frequency responsiveness. In order to combine the benefits of FL and conventional PID control, the proposed LFC employs a two-stage control mechanism. The unpredictable variations in wind output produce nonlinear conditions in the control area of the WT generator, which can be adequately compensated for by the fuzzy LFC. According to the simulation results, FGS-PID LFC operates better in a control area with low inertia [172].

To maintain frequency and voltage stability in microgrids, Angalaeswari et al. suggest an iterative learning controller (ILC) that can function in fluctuating situations in an autonomous grid. The suggested ILC outperformed competitors in both off-grid and on-grid modes under a variety of conditions for voltage and frequency stability. Voltage imbalance is minimized and running duration is decreased with this method. The suggested ILC achieves a maximum error reduction of 79% [173]. Thao et al. propose a FL-based two-stage framework. The goal of this coordinated FL structure is to manage the grid frequency by adjusting the active power provided to the grid by the PV plant. Finally, simulations reveal that the suggested technique can regulate frequency in transient and steady-state modes within permitted deviation ranges of 0.2 Hz and 0.05 Hz, respectively [174]. To optimize the operating efficiency of DC side voltage controllers in grid-connected PV systems, Tripathi et al. suggest a new method, termed fictitious reference iterative tuning. In the investigation, 2-DOF PI is employed, and DC side control is achieved with enhanced distortion response. As a result, it was established that improved results in DC voltage regulation were obtained [170]. To improve the frequency response of doubly-fed induction generators (DFIGs) in wind energy conversion systems, Peng et al. suggest a coordinated FL-based algorithm using the DFIG-energy storage couple.

The system’s frequency responsiveness is enhanced with the suggested design in the short term and for various wind speeds [175]. In a different study, Peng et al. used a FL-based algorithm to execute optimal control strategy and sizing with a DFIG energy storage pair in wind energy conversion systems. According to the claim made in [176], the suggested system performs properly in terms of frequency regulation and power generation in WT and battery storage systems. A smart coordinated control is proposed by Roselyn et al. In order to reduce the imbalance between supply and demand while the microgrid is operating, this method, which is based on ANFIS, combines control techniques such as current, virtual impedance, and reverse fall controls. Experiments based on simulation have been conducted to compare the fast switching response of different control methods. The proposed design is said to perform better in various system situations as a result of investigations [177]. To manage frequency oscillation in power networks, Yu et al. suggest an adaptive multiple-input and single-output (MISO) FL algorithm. They want to make the system less expensive. In the controller design, the PSO algorithm is utilized for the system’s effective operation. It was concluded that the MISO structure functions more effectively in comparison to other structures as a result [178].

In order to reduce the detrimental impact of partial shadowing situations on the frequency regulation of the grid in big power PV systems, Rahman et al. suggest a new ANN-based control technique. The study’s findings demonstrate that the suggested technique has a favorable impact on the network’s frequency performance [179]. A control system based on ANNs was suggested by Kaushal et al. By controlling power quality in accordance with IEEE/IEC standards, this technology reduces voltage and frequency fluctuations in systems that connect distributed resources. In the simulation environment, a microgrid structure was created using the proposed controller structure [180]. In order to estimate very short-term frequency voltage in renewable energy systems, Chettibi et al. suggested a technique based on ANN and deep recurrent neural networks. The University of Manchester’s on-grid battery energy storage system was used to provide the validation data that were used to test the system. It was found, as a result, that the created algorithm predicts the variables with a reasonable level of accuracy [181]. According to the literature review cited above, using FL and ANN applications to stabilize the grid’s voltage and frequency is a very important strategy.

4.2.2. Applications of Fuzzy and Ann in Active Reactive Power Quality Control

Different techniques have been used to implement FL and ANN algorithms for PQ control and power quality improvement of grid-interactive systems that operate concurrently with RES. The studies in the literature that specifically use FL and ANN are listed below. Shunt hybrid filters were used in a design by Das et al. to enhance power quality and PQ in microgrids. The proposed architecture employed a fuzzy neural network, and the outcomes were evaluated in comparison to those of alternative control strategies [182]. In order to improve the power quality of a solar energy conversion system (SECS), correct the compensation for reactive power, and maximize active power generation, Ouai et al. suggested a model for the potential utilization of PV plants and for power conditioning to operate at full capacity. The study’s findings demonstrate that the proposed control technique performs optimally and at its highest level of efficacy. It confirms that SECS can be completed at its maximum power, even though the power quality in the planned system can be enhanced with active filtering and reactive power adjustment [183].

Hou et al. presented a fuzzy proportional complex integral control method to reduce the quasi-Z source on-grid PV inverter’s negative control effect. For the AC distortion signal of a chosen particular frequency and excluding the steady-state error, this approach includes the exclusive of zero steady-state error improvement [184]. Using a supercapacitor as energy storage, Carvalho et al. suggested a FL-based model for power smoothing of WT power fluctuations. The supercapacitor energy storage’s state of charge (SOC) is controlled by the model both during successful and unsuccessful WT power smoothing and in the simulated microgrid. In all circumstances, the proposed model exhibits improved and good performance of the FL-based method compared to the baseline scenario and conventional smoothing technique [185]. Faroug et al. created proportional–integral (PI) controlled static synchronous compensator (STATCOM) controllers. After this, STATCOM controllers with FL control were compared. By introducing reactive electricity into the grid, STATCOM and fuzzy logic-controlled STATCOM corrected the problem. By preventing the tripping of WT generators, they used to be able to stabilize the voltage in wind turbine (WT) power [186].

For an effective low-voltage ride-through (LVRT) capability, Roselyn et al. developed a smart fuzzy-based reactive and real power regulation of the inverter. The suggested LVRT solution [187] incorporates a dynamic braking resistor and a braking chopper to categorize the overshoot in DC connect voltage during grid errors. Babu et al. suggested a fuzzy-based ideal point speed ratio MPPT controller for an on-grid wind system. An FLC, a PWM inverter, a DC boost chopper, a three-phase uncontrolled rectifier, a permanent magnet synchronous generator, and a three-bladed permanent field turbine are all components of the overall system. The MPPT controller on the generator side succeeded in obtaining the WT’s maximum power at any speed [188]. With the aid of a battery energy storage system, Yang et al. optimized fuzzy logic empirical mode disintegration to lessen the variations of WT power. For this reason, the WT power signal was divided into two parts: the low-frequency part and the high-frequency part. The battery system’s level of charge and the pace of power fluctuation are advanced restrictions that the filter is proposing. Results show that the proposed model can perform passable smoothing of wind power variation [189]. Efrain et al. made reactive FL-based control of smart PV inverters possible.

In this study, smart distribution power grids were simulated using a real-time digital simulator. Results show that the suggested model lowers power losses, lowers PV reactive power injection synchronously, and increases magnitude voltage [190]. A distributed power flow controller-based integrated hybrid system model was created by Naidu et al. FOPID and PQ theory controllers are employed in this hybrid model. ANFIS, FUZZY, and PI controller comparisons are used to validate the proposed system. As a result, the FOPID controller performs better in terms of harmonic reduction, transition fall and rise voltages, and voltage compensation [191]. Rezaie et al. used a static fuzzy-based volt-ampere reactive (VAR) compensation to increase a wind turbine system’s capacity to operate at low voltage. The proposed controller would reduce the maximum voltage drop during fault conditions and increase steady-state voltage. The biggest voltage losses occur after the steady-state voltage error at the load bus and the fault incidence at the WT generator bus, which are reduced by an average of 25.16% and 74.44%, respectively [192].

In a PV system connected with a microgrid, Kaushal et al. developed power quality control based on power factor, total harmonic distortion, frequency, unbalancing, and voltage sag/swell using ANN. This system’s effectiveness is demonstrated by the line communication and impedance delay development for the evaluation of power quality metrics. As a result, the suggested controller can function as intended even if RES’s nature changes [180]. In order to solve the power quality issues, Kumar et al. used a WT, PV, and fuel cell-based hybrid renewable energy system. Some authors used ANFIS, ANN, and FL-based control algorithms to maximize the dynamic performance of this filter. These controllers employ nonlinear, balanced, and unbalanced loads to achieve the smooth DC link voltage while minimizing the total harmonic distortion produced in [193]. An ANN-based direct power control technique was made possible by Douiri et al. to synchronize DFIG and manage power flow with voltage-oriented control and a grid. The suggested model architecture achieves significantly shorter performance times, and the faults that cause time delays are decreased [194]. In order to optimize the power output of a small-scale Darrieus vertical axis wind turbine, Abdalrahman et. developed intelligent blade pitch control. Dynamic ANNs were utilized by Mordjaoui et al. to forecast daily power consumption. On the basis of load data, the effectiveness and applicability of this procedure were established. This information came from the website of the French transmission system operator. Results of the forecasting approach show that for historical and predicted loads, respectively, the mean absolute percent error is 4.223% and 3.266% [195]. Seven alternative ANN training methods were examined by Monteiro et al. to estimate the generation of active power. Researchers compared the performance of the Kalman filter and support vector machine in this study using data on the number of hours in the day, solar panel temperature, air temperature, and irradiance. Therefore, the relative MAPE results for ANN, SVM, and KF are 0.02%, 0.33%, and 3.41% [196]. Saviozzi et al. used ensembles of ANNs for load prediction and modeling. The data set for this investigation included historical load time series and data associated with network users. Results show that using both of the suggested strategies can help people achieve successful outcomes on actual distribution networks [197]. In order to successfully implement PQ power regulation in the grid, the literature review described above emphasizes the importance of FL and ANN applications. The summary of evaluation of different optimization AI techniques in microgrids is shown in

Table 3. Summary of different optimization AI techniques used in microgrids.

Ref	AI Technique	Objectives/Contributions	Method/Mode
[198]	ANN	Generation capacity optimization	Simulation
[199]	ANN and cooperative control	Voltage and frequency regulation	Simulation
[200]	Distributed ANN	Energy management system	Real-time experiment and simulation
[168]	GA	Hybrid PV-wind and battery storage	Coded simulation
[201]	PSO	Hybrid SPV and WTG; scattering and optimization	Simulation
[5]	FL	Hybrid SPV and WTG; input–output data	Simulation
[202]	PSO	Minimization of costs for various MGs, including RESs; expenses for operation, emissions, and MG dependability are minimized.	Islanded
[203]	GA	Operating costs, discharge costs, and power exchange benefit are objectives in a multiobjective EMS design for the best performance of MG.	Both
[204]	ABC	MG domestic operating costs are constrained by a two-layer control design that has received preliminary clearance.	Grid-connected

.

5. Review of Applications of Artificial Intelligence (AI) Strategies under Fault Outbreaks in Microgrids

ANN, FL, PSO, GA, and PSO are just a few of the AI-based fault detection and classification methods that have recently been used for microgrid protection. For a microgrid protection system to enable speedy repairs—including network restoration—and cut down on interruption time, efficient fault detection and its categorization methodologies are necessary. For quick and accurate fault detection and classification for the microgrid system, however, it is necessary to take into account the performance of AI-based techniques and their computational time. In numerous investigations, various AI-based defect detection and classification approaches have been published and summarized in this paper.

5.1. Energy Flow and Management under Faults

In order to ensure stable, dependable, and sustainable operation of the microgrid and other operational goals such as minimizing costs and fuel consumption [205,206], an energy management system (EMS) is important [207], it is a control tool that manages the power flows among the main grid, DERs, and loads. Additionally, it is in charge of resynchronizing the system while switching between grid and island mode. Decentralized and centralized control, with varied hierarchical controls, are the two basic EMS strategies that can be distinguished [208].

5.1.1. Centralized Controller

All the measured data from all DERs in the microgrid are gathered by the centralized controller, which subsequently modifies the control variable for each piece of equipment and sends the updated data to the central system [209]. In particular, small-scale MGs are a perfect fit for this control. The dependability and redundancy of this form of control are low. The fact that this control may result in a number of communication issues and that it necessitates system shutdown in the event of system maintenance are two additional downsides. Centralized hierarchical control offers a successful answer from an economic perspective. Depending on the type of microgrid or the existing infrastructure, the hierarchical control architecture may be used. The following three tiers of controllers could make up a centralized hierarchical control scheme in this situation: local controllers, a microgrid central controller (MGCC), and a distribution management system (DMS).

Because communication networks are frequently avoided for reasons of reliability, local controllers employ local measurements to manage the voltages and frequency of the MG system without the need of communication systems. Each microgrid has an MGCC that can connect to the DMS. The MGCC manages the microgrid’s power by assessing the DERs’ active power, load demand, and storage needs. The two-way communication between the MGCC and LC allows it to fulfill the utility requirements [210]. Overall grid demands and stability requirements are satisfied at the DMS level [211].

5.1.2. Decentralized Controller

In a decentralized controller, independent control of DER units and loads is possible because of the decentralized EMS. If microgrid users have different goals or operate in a different operational environment, this sort of EMS is more appropriate. All local controllers in this management system are linked together by a communication bus. Each household or DG exchanges data using this bus [206]. In such a distributed system, local controllers are no longer subject to an MGCC to choose the best power output. As a result, the computational demand is greatly reduced by this type of structure, which also relieves pressure on the communication network throughout the entire microgrid system [205]. The most effective illustration of a decentralized energy management system is the multiagent system (MAS) approach [212]. This technology uses artificial intelligence-based techniques such as neural networks and fuzzy systems to establish each DG’s operation point while enhancing the stability of the microgrid [206]. Its goal is to break down big, complicated systems into smaller, autonomous subsystems. Compared to centralized EMS, the decentralized-based MAS has a number of benefits. The MAS reduces calculation time since it allows DGs to work autonomously and uses the necessary data from the local controller. However, centralized control necessitates a sizable data flow to a single location [213,214]. The flexibility to plug and play is another benefit. A programmable agent in its control is supplied without changing the rest of the control if a new DER is linked to the microgrid. However, when a new DER is connected under centralized controls, the MGCC must be programmed [214].

Table 4. AI-based control for fault dynamic response and power quality enhancement of MGs summary.

Ref	AI Technique	Objectives	Control Strategy/DGs	Grid Connect (On/Off)
[215]	ANN	Power sharing; voltage regulation	Centralized	Off
[216]	ANN	Voltage and frequency regulation	Centralized	Off
[217]	ANN	Frequency regulation	Centralized	Off
[218]	ANN	Power-sharing droop control	Centralized	Off
[219]	ANN	Voltage stability; power sharing	Decentralized	Off
[220]	SLFN	Communication delay	Decentralized	On
[221]	ANN	Optimal control	Distributed	On
[222]	ANN	Power quality	Distributed	On
[223]	PSO	Regulation of voltage and frequency; enhancement of dynamic response	Distributed	Off
[224]	PSO	Voltage and frequency control and compensation of reactive power	Distributed	Off
[225]	PSO	Regulation of active and reactive power	Distributed/wind solar PV	On
[226]	PSO	Transient response improvement	Distributed solar PV	On
[227]	PSO	Harmonic modification and power factor enhancement	Distributed/wind-PV fuel cell, diesel	Both
[228]	PSO	Voltage stability enhancement	Distributed/ wind-PV	Both

presents AI based control for the dynamic response and power quality enhancement of MGs.

6. Conclusions and Future Outlook

An overview of AI-based control in microgrid contexts is given in this study. The necessity for AI approaches and their implementation at the various levels has been examined and future scopes have been presented. An overview of current traditional control methods, optimization strategies, fault circumstances, and their limitations have also been evaluated. Despite the difficulties and complexity of system models, it is discovered that AI can unquestionably be a crucial instrument for facilitating the smooth integration and control of DERs at the local and networked levels. On all levels, ANN-based models are receiving increased attention. The majority of AI techniques currently in use are simulation and physical model-based but data-driven techniques should also draw more attention. The design becomes less difficult since a data-driven model does not need comprehensive physical system information. There is a deficit of data at the level of the low-voltage distribution network but for some particular applications, such as inertia calculation in the primary control. The public data sets for ESS levels need considerable preprocessing to improve the data quality and predictability because ESS is becoming a crucial component of decarbonizing smart and microgrid networks.

The microgrid domain does not currently use semisupervised learning, although it can be applied in cases where the supplied data set is not entirely labeled. Another barrier to efficient control in the secondary layer is the communication infrastructure. While further study is needed to lessen the complexity of model construction, the current research on AI-based communication focuses on ANN-based strategies that have been adopted to mitigate the communication requirement. Autonomous market participation is another area in the tertiary layer that needs more study. PSO is frequently used in microgrids for control and optimization. Research on FL and GA is ongoing as well. In networked microgrids, several AI-based solutions have not yet been taken into consideration. The validation of the created/proposed solutions for standalone or networked microgrid contexts needs to be a priority as well. The conventional power system approaches create constraints in processing and analyzing the vast volumes of data that are now the standard with a smart grid as the traditional electric grid system evolves into a smart grid system. In order to address a variety of applications in smart grid systems, AI approaches are being developed and put to use, and the results thus far are encouraging.

In addition, this work gives a summary of new applications of AI techniques in four crucial domains (i.e., optimization, load forecasting, power grid stability assessment, defects detection and security challenges) not previously covered in earlier research. The use of AI approaches to build a genuinely smart grid is also covered, along with current difficulties, potential gains, and future prospects. The study can be summed up as follows based on the results of this survey: (i) AI techniques have been used in a number of application areas essential to the dependability and resilience of a smart grid. (ii) Despite this, there are still some obstacles preventing further applications of AI techniques. The handling of the fault’s nature of some AI techniques to achieve a human-centered approach to AI solution design, as well as data privacy and security, are among the biggest challenges. (iii) This survey should stimulate discussions in the application areas surveyed in this paper, which could further strengthen exchange of ideas. In conclusion, smart grid systems’ reliability and resilience are enhanced and improved through the use of applications of AI approaches. Future research in this field will concentrate on examining the effects of the AI approaches character on smart grid operations and develop new optimization formation strategies. Future study will look at how this issue has been addressed by smart grid operators. Such a poll could assist academics in developing AI solutions that are more focused on the needs of people.