Reduction in Residential Electricity Bill and Carbon Dioxide Emission through Renewable Energy Integration Using an Adaptive Feed-Forward Neural Network System and MPPT Technique

Abstract

Increasing electricity demand and the emergence of smart grids have given home energy management systems new potential. This research investigates the use of an artificial neural network algorithm for a home energy management system. The system keeps track of and organizes the use of electrical appliances in a typical home with the objective of lowering consumer electricity bills. An artificial-neural-network-based maximum-power-point-tracking scheme is applied to maximize power generation from photovoltaic sources. The proposed neural network senses solar energy and calculates load requirements to switch between solar and grid sources effectively. The implementation of improved source utility does not require numerical calculations. Traditional relational operator techniques and fuzzy logic controllers are compared with the suggested neural network. The model is simulated in MATLAB, and the results show that the artificial neural network performs better in terms of source switching following load demand, with an operating time of less than 2 s and a reduced error of 0.05%. The suggested strategy reduces electricity costs without affecting consumer satisfaction and contributes to environmental friendliness by reducing CO₂ emissions.

1. Introduction

In the past decade, the electrical energy generation sector has faced a critical issue due to increased energy demand, fossil fuel depletion, and environmental pollution due to the emission of ${\mathrm{C}\mathrm{O}}_2$ [1]. In line with the United Nations 26th Conference of Parties (COP 26) held at Glasgow in November 2021, India aims to attain 500 GW of non-fossil energy capacity by 2030 and to reduce carbon intensity by 45% to achieve net zero emissions by 2070 since India is the world’s fourth highest ${\mathrm{C}\mathrm{O}}_2$-emitting country, with a total ${\mathrm{C}\mathrm{O}}_2$ emission value of 2.299 million tonnes per year [2]. Therefore, electricity generation from renewable energy sources has gained significant importance. Integrating renewable energy sources, such as photovoltaic and wind turbines, into the power grid transitions the power system from a centralized network to a decentralized, bidirectional electrical network. According to India’s Ministry of Powers report, non-renewable energy sources account for 64% of total energy production [3]. A Ministry of New and Renewable Energy Sources report states that India attained 104.88 GW of renewable energy capacity in December 2021. Sector-wise installation of renewable energy sources in India by 2021 is shown in

Table 1. Renewable sector installation in India by 2021.

Sector	Target 2022 (GW)	Installed (GW)	Presently Being Implemented (GW)
Wind Power	60	40.08	9.65
Solar Energy	100	49.35	40.86
Small Hydro	5	4.84	0.36
Bio Energy	10	10.61	0
Thermal + RE Building (Hybrid)	0	0	5.44

. The annual energy outlook with projections to 2035 states that an increase in the world population will lead to a 24% increase in energy demand by 2035 [4]. Residential energy consumption is the third highest among all the economic sectors that consume energy, including the transportation, industrial, and commercial sectors [5]. To motivate electricity customers to actively participate in load management, smart grids and utilities worldwide have proposed various incentives and initiatives, such as time of use rates, demand response (DR), and real-time pricing [6,7]. Figure 1 shows different schemes that have been proposed for integrating renewable energy sources in India by 2021.

Requirements to maintain the energy consumption rate and improve the reliability and efficiency of energy usage have led to the introduction of energy management systems (EMSs). An EMS is a computerized tool used to control, monitor and optimize the performance of residential electric power [8]. Home energy management systems (HEMSs) have been introduced to facilitate automatic energy management in the home. HEMSs reduce energy demand by shifting the peak hour’s load to non-peak hours, enhancing household convenience [9]. A DR program presented by the load-sharing entities helps to improve active end-user participation. DR programs are categorized as incentive-based and price-based programs [10]. Researchers have introduced numerous techniques for implementing HEMSs based on demand response programs and load scheduling. A heuristic-algorithm-based mixed scheduling technique [11] approaches the desired results. This method suggests a complex consumer process: fixed and irregular appliance scheduling. Due to this technique, customers find it uncomfortable and the installation and maintenance cost is high. A profile-matching shift optimization algorithm [12] for home energy management systems makes decisions based on a profile matching algorithm, which includes deferrable and non-deferrable loads based on real-time tariffs. Obtaining household appliance operation times negatively impacts how well customers are served. The time-of-use pricing method operates based on the forecasting technique. This method involves dividing the individual household energy consumption into three states for monitoring: forecasting, day-ahead planning and accurate operation method [13]. Complication in prediction occurs due to inaccurate assumption values. The multi-objective optimization algorithm suggests a home energy management system based on load scheduling [14]. Although home energy consumption has a significant impact on environmental degradation, this method does not take such effects into account.

ANN-based HEMS [15] with a smart plug and PV production approach is recommended for the smooth operation of the system. The decision made by the system is solely based on the appliance running times. Because home appliances work based on their environment and users’ values, each has a specific running time, making consumers uncomfortable. The home energy management strategy suggested by neural network technology takes into consideration appliances that are left on. This method considers the specific functioning region of the appliances under various circumstances [16]. With the suggested ANN-based method, it is difficult to formalize the pattern of appliances and the scheduling process is unclear. Therefore, customers could not apply this strategy to all household gadgets. As a result, the consumers’ quality of experience suffers [17]. Model-based predictive control and the branch-and-bound technique [18] propose an HEMS based on appliance scheduling. This methodology considers the energy consumption and cost reduction process. The suggested method does not consider the consumer experience in terms of quality and environmental effects. The metaheuristic genetic algorithm [19] based HEMS methodology suggests HVAC-type appliance scheduling based on the tariff and operating time. Therefore, this suggested method is unsuitable for all home appliances, such as those with low power consumption. The implementation of this method is complicated in the residential sector.

The performance of neural networks [20] determines the procedure applied to predict how much energy would be used by commercial buildings in various weather scenarios. Only the prediction of energy usage is performed using this method. The suggested ANN [21] algorithm is used in a hybrid solar and wind energy management system. However, this system does not consider the appliances’ status and operating methods. The suggested ANN method is appropriate for a single residential building, but community-level implementation is difficult in terms of data collection and execution [22]. To meet the demand for electricity in the residential sector, artificial neural networks and fuzzy systems will regulate the power supply. The load scheduling and user experience quality are not made explicit by this approach [23]. The machine learning–backpropagation algorithm [24] technique proposes switching mechanisms and the controller decision is purely dependent on the trained data. This method collects the data on various parameters from residential users. In this method, it is not clear which type of appliances are used in the system.

Analyzing the various proposed schemes, it is observed that load scheduling is linked to real-time electricity tariffs. Load scheduling techniques that provide the expected results do not consider the customer’s quality of experiences and the environmental impacts. Hence, they require additional cost-reduction methodologies to enhance the quality. Introducing new household appliances, such as battery energy storage, air conditioning, electric vehicle charging and so on, makes it difficult for customers to calculate the load manually. The data collection step is crucial for HEMS and for data collection, researchers use a sensor and meter with value prediction methodology. This value prediction process is based on primary data (previous data) and applies to renewable energy sources (RES) and load prediction. Prediction techniques produce better results; even though the value is assumed, highly qualified methodologies for analysis and reducing prediction error are required. Numerous prediction techniques based on machine learning algorithms have also been proposed, but they possess certain limitations. A fuzzy-logic-and-wireless-sensor-based smart thermostat for residential energy management was presented [13], but there is significant complexity in its real-time implementation. In [12], the researcher proposed power-scheduling-based heuristic algorithms, including fuzzy and artificial neural networks (ANN). However, the model is limited to home automation. In [25], an ANFIS-based HEMS is proposed, but this approach needs to include the integration of renewable energy sources with the grid. Integration of renewable energy sources with the smart grid also plays a significant role in the home energy management system, but this approach suffers from demand matching [14,26]. Integrating renewable sources into the grid reduces demand on the grid and enhances its reliable operation [27]. Among all the renewable energy sources, photovoltaic power generation has gained considerable importance in HEMSs in recent years, because it can be installed on the rooftop of a single house and provide humanity with clean, climate-friendly, abundant and inexhaustible energy resources [28,29,30]. The output power from the solar panel will fluctuate in nature since it is affected by environmental conditions such as irradiance and temperature. The forecasting of its effects on the electrical grid and the execution of the EMS, which depends on the forecasting of PV power generation, are made unclear by these uncontrollable factors. This makes accurate forecasting of PV power output a significant obstacle to solve to accomplish massive PV integration, which has sparked numerous studies on a global scale [31].

2. Motivations and Contribution

Through literature surveys, several major issues were identified regarding the integration of solar power with the grid for home energy management (HEM). These issues include ${\mathrm{C}\mathrm{O}}_2$ emissions, increased electricity bills and inconsistent solar output due to varying temperatures and irradiance. The proposed method aims to address these issues and can potentially reduce the cost of electricity that consumers pay by implementing grid-tied solar photovoltaic systems in their homes. This serves as a motivation for the proposed work.

The contributions of the proposed work can be summarized as follows:

• Analysis of Different Approaches: The proposed method involves analyzing three approaches: the relational operator approach, fuzzy logic controller and artificial neural network approach. Through a detailed analysis, it is determined that the artificial neural network approach is the most effective one. This analysis contributes to the field by providing insights into the suitability and effectiveness of various approaches.
• Performance Improvement of HEMS: The proposed work aims to improve the performance of the HEMS by implementing model-based predictive control algorithms with improved forecasting capabilities. This contribution enhances the efficiency and effectiveness of HEMS, addressing the identified issues and promoting a more sustainable energy management system.
• Maximum Power Point Tracking (MPPT) Technique: The proposed method includes the implementation of an MPPT technique using a suitable machine learning algorithm. This technique enables the solar panel to extract the maximum power available under varying temperatures and irradiance conditions. Applying this technique reduces ${\mathrm{C}\mathrm{O}}_2$ emissions from the domestic sector, promoting environmentally friendly energy generation.
• Consumer’s Quality of Experience: Recognizing the importance of the consumer’s quality of experience in energy management, the proposed system ensures a positive and satisfactory experience. This contribution focuses on meeting the needs and preferences of consumers, enhancing their overall satisfaction and comfort.

Based on the identified problems, our proposed system uses a grid-tied process. Home loads are segregated based on the priorities of consumers This method is an affordable process for consumers because the government provides a subsidy for solar panel installation, as shown in

Table 2. Subsidy details for PV installation.

Description	Cost (INR)
Cost of 1KW rooftop solar system	46,923
Subsidy @ 40%	18,769
Cost after subsidy	28,154
Net cost after subsidy	28,154

.

Organization of This Paper

This paper is structured in a way that includes five sections, including an introduction. The design of the solar panel and the implementation of an ANN-based MPPT scheme are discussed in Section 3; a proposed methodology that includes the relational operator method, fuzzy logic and ANN is discussed in Section 4; the simulation and results of the proposed system are discussed in Section 5, followed by conclusions and future research in Section 6.

3. Photovoltaic Installations

3.1. PV Cell

The PV cell is made of materials such as silicon (Si), gallium arsenide (GaAs), copper indium diselenide (CIS) and cadmium telluride (CdTe) [32]. The mostly equivalent circuit of a PV cell is represented as a PN or Schottky junction that enables the PV effect. A single solar cell is not capable of supplying the desired voltage. Hence, they are connected in series and parallel to constitute a PV module/array [33]. The requirement of increased voltage or increased current decides the connection of solar cells or modules in parallel or series. The classification of solar cells with their operation efficiency is shown in

Table 3. Solar cell efficiency based on cell construction.

Solar Cell Type	Benefits/Efficiency of Solar Cell	Expensive
Mono-crystalline silicon	Efficiency—15–20% It can provide better performance in low-light conditions. Better lifetime	Most expensive
Poly-crystalline silicon	Efficiency—13–16%	Better economic choice
Thin-film solar panels Cadmium telluride cells (CdTe) Copper indium gallium diselenide (CIGS) Amorphous silicon (a-Si), thin-film silicon (TF-Si)	Efficiency—6–11% Thin-film cells are used in low-voltage applications	Less expensive, less efficiency
Hybrid silicon solar panels	Maximum efficiency—16%, Mix of amorphous and mono-crystalline cells	Most expensive

.

3.2. Mathematical Model of PV Cell

Initially, the ideal single solar cell model is designed with a current source, resistor and diode. Later, the procedure is adopted for the design of the PV module [34]. The equivalent circuit of a solar cell is shown in Figure 2, which consists of a diode, current source, series resistor (R_s) to represent internal resistance that resists current flow and shunt resistor (R_sh) to express leakage current [35,36,37]. By the application of the Shockley solar cell equation, the I-V characteristic equation can be given as follows:

$$I=I_L-I_D$$

(1)

$$I=I_L-I_0 (e^{\frac{qV}{kT}-1})$$

(2)

where $I$ = output current (A), $I_L$ = photo-generated current (A), I_D = diode current (A), and V = terminal voltage. In practical terms, the V-I characteristics of the solar cell include shunt and series resistance. Therefore, Equation (1) can be rewritten as follows:

$$I=I_L-I_D-I_{SH}$$

(3)

$$I=I_L-I_O (e^{\frac{qV}{kT}}-1)-I_{SH}$$

(4)

$$I_{SH}=\frac{V_{SH}}{R_{SH}}=\frac{V_D}{R_{SH}}=\frac{V+I_{RS}}{R_{SH}}$$

(5)

Substituting Equation (5) in (4), we obtain

$$I=I_L-I_0 {(e}^{\frac{qV}{kT}}-1) - \frac{V+I_{RS}}{R_{SH}}$$

(6)

To enhance the power output of the solar cells, numerous cells are connected in series and parallel. Hence, the output current equation of a solar module consisting of multiple series and parallel connected cells is given in Equation (7). Ideally, the solar cell Rs is very small (Rs = 0) and R_SH is very large (R_SH = α). Then, the output current equation will be as depicted in Equation (8).

$$I=N_P{I}_L-{N_{P}I}_0\left(exp\left(\frac{\frac{V}{N_S}+\frac{I{R}_S}{N_P}}{\frac{nkT}{q}}\right)-1\right)-\frac{\frac{N_P}{N_S}V+{IR}_S}{R_{SH}}$$

(7)

$$I=N_P{I}_L-{N_{P}I}_0\left(exp\left(\frac{\frac{V}{N_S}}{\frac{nkT}{q}}\right) 1\right)$$

(8)

where $N_P$ is the number of parallel connected cells, N_S is the number of series connected cells and n is the identity factor that depends on the material used in the PV technology; for example, the n values of CdTe, Si-poly and CIS are 1.5, 1.3 and 1.5, respectively.

3.3. MPPT Using Artificial Neural Network

Solar radiation and environmental temperature are the crucial factors that define solar output. Stability in solar output is necessary to integrate it with the grid. However, this presents real-world challenges. This paper proposes an artificial-neural-network-based solar maximum power prediction method to reduce the computational cost of real-time prediction systems and achieve high prediction performance. Artificial neural networks are machine learning algorithms that replicate the function of the human brain and find the solution for the most challenging issues. Neurons in the network handle the collection of data and the output predictions. In the proposed work, temperature (T) and radiation (G) are given as the inputs and the output is maximum power. The block diagram and schematic structure of the proposed network are shown in Figure 3 and Figure 4.

Among the various training methods, the back-propagation approach with the Levenberg–Marquardt algorithm is selected for the proposed work. In total, 1000 data sets were generated per the pseudo-code in Figure 5. Among these data, 70% are used for training and 30% are used for testing and validation. The main objective of the ANN is to obtain the best prediction accuracy by reducing the root mean square error (RMSE). The reference PV voltage is compared with the output voltage produced by the ANN based on the temperature and irradiance obtained. The difference in the voltage is fed to the preoperational integral derivative controller to generate the control signal, as shown in Figure 6. The control signal generated by the control unit based on the ANN output varies the converter’s duty cycle to obtain maximum power. The output of the converter is impacted by changes in the duty ratio. Therefore, the suggested ANN-based MPPT technique can precisely detect the duty cycle under a range of solar irradiation [38]. The normalized duty cycle solar radiation and temperature conditions are displayed in

Table 4. Basic operation of ANN-based MPPT control.

Voltage Derivation (dv)	Power Derivation (dp)	dp/dv	Duty Cycle
−1	−1	+1	D(n) = D (n − 1)
+1	−1	−1	D(n) = D (n − 1)
−1	+1	−1	D(n) = D (n − 1)
+1	+1	+1	D(n) = D (n − 1)

. Figure 7 and Figure 8 show the solar output based on the MPPT technique.

The relationship between the duty cycle and the voltage and current at the maximum power point (${\mathrm{V}}_{\mathrm{m}\mathrm{p}\mathrm{p}}$,${\mathrm{I}}_{\mathrm{m}\mathrm{p}\mathrm{p}}$) is given as follows:

$$D=1-\sqrt{\frac{{\mathrm{V}}_{\mathrm{m}\mathrm{p}\mathrm{p}}\times {\mathrm{I}}_{\mathrm{o}\mathrm{u}\mathrm{t}}}{{\mathrm{I}}_{\mathrm{m}\mathrm{p}\mathrm{p}}\times {\mathrm{V}}_{\mathrm{o}\mathrm{u}\mathrm{t}}}}$$

(9)

4. Proposed Technology

A block diagram of the proposed system is shown in Figure 9, which suggests the implementation of the HEMS for an individual home with grid-tied solar PV. PV systems connected to the grid can export and import electricity as needed. The main components of the proposed system include a grid-tied PV system, an inverter and a power conditioning unit. Since most household appliances are powered by alternate current (AC), the PV system generates direct current (DC) electricity, which must be converted to AC. As a result, with a grid-connected PV system, the PV inverter is essential. All extra electricity in a grid-interactive system is fed into the grid.

Since the HEMSs are implemented with RES integration in the current method, the home appliances are divided into solar-connected and grid-connected categories. Solar-connected appliances are directly connected to solar energy via the inverter battery, whereas grid-connected appliances are directly connected to the power grid. ANN-based MPP extraction is carried out. The relational operator and fuzzy-logic- and neural-network-based controllers are used to perform source switching without disturbing customer satisfaction.

The conditions preferred during execution are as follows:

• During an emergency, at least one priority load, namely a solar-connected load, must be operated.
• It achieves energy conservation by transferring excess solar energy to the grid.
• During no-radiation conditions, the grid will feed the solar-connected loads.

5. Simulation and Results

The proposed methodology is simulated in MATLAB R2022a software. The simulation connects residential loads to solar power, while the remaining loads are connected to grid power. The primary goal is to switch loads between solar and grid power based on solar power availability. The solar output power is determined by irradiance and temperature values. The solar panel under consideration for analysis is the Solar Tech, ASC-6P-36–130 Watts. Temperature and radiance are two input variables used to analyze the proposed system. These two variables are related to meteorological parameters such as ambient temperature (T), humidity level (RH), wind speed (WS), cloud cover (CC) and precipitation (P).

Table 5. Monthly solar energy availability.

Month	Global Irradiation kWh/m2	Available Solar Energy kWh
January	178.8	316.7
February	178.8	296.0
March	201.9	306.8
April	189.6	271.0
May	185.4	255.4
June	134.1	186.5
July	135.6	192.6
August	154.1	223.5
September	153.2	231.0
October	157.4	252.1
November	141.0	239.4
December	172.9	312.5

shows the monthly solar energy availability in a solar site located in Coimbatore, Tamil Nadu, India (latitude—11.01° N, longitude—76.97° E). The solar panel inclination angle is 11° and the altitude is approximately 442 m. The prediction process is performed through the PVsyst simulation tool based on the geographical value.

The module consists of one parallel string and ten series-connected modules per string. The parameters of the considered solar panel are listed in

Table 6. ASC-6P-36-130 watts solar panel parameter.

Description	Value
Maximum power	129.9672 W
Open circuit voltage,$V_{OC}$	21.69 V
Voltage at maximum power point, $V_{mp}$	16.41 V
Short-circuit current, $I_{sc}$	8.42 A
Current at maximum power point,$I_{mp}$	7.92 A
Temperature coefficient of VOC	−0.35998 °C
Temperature coefficient	0.06 °C
Shunt resistance, $R_{sh}$	153.497 Ω
Series resistance,${ R}_s$	0.35008 Ω
Diode ideality factor	0.97321

. The current voltage (I–V) and power voltage (P–V) characteristics of the selected solar panel under various dynamic operating conditions are shown in Figure 10, which describes the voltage from solar to load after implementing the ANN-based MPPT scheme. The voltage in the solar panel varies as per the irradiance and temperature. The oscillation in the converter and inverter is due to the initial test analysis of sudden variation in solar output. Hence, to avoid oscillation of the inverter output, a single-phase inverter with a pulse width modulation technique is employed in the proportional method. The inverter output voltage will be based on the solar voltage and the switching sequence of the inverter. The objective of the proposed system is to minimize the total electricity cost.

5.1. Relational-Operator-Based Source Switching

In this methodology, home appliances are categorized into three categories based on consumer needs: (i) RES-connected load, (ii) grid-connected load and (iii) RES/grid-connected load. RES-connected appliances are connected to the RES (PV) and grid-connected devices are connected to the grid supply. The RES/grid-connected appliances can operate in the source-switching mode throughout their operation, as shown in Equations (10) and (11). Usually, the RES/grid-connected appliance is connected to the PV system if the PV output is insufficient to run the total load demand. At that time, RES/grid-connected appliances can switch to the grid source. The proposed switching can be done through the relation operator to achieve this. This reverse process is based on PV output appliances using either PV or grid power.

$$\sum _{i=0}^n{\mathrm{X}}_{\mathrm{i}}+\mathrm{R}<Y$$

(10)

$$\mathrm{If} {\sum }_{i=0}^n{\mathrm{X}}_{\mathrm{i}}+\mathrm{R}>Y \mathrm{then} R=Z$$

(11)

where ${\mathrm{X}}_{\mathrm{i}}$—solar-connected appliances $({\mathrm{X}}_1,{\mathrm{X}}_2\dots {\mathrm{X}}_{\mathrm{n}})$, R—solar/grid-connected appliances, $Y$—solar supply and $Z$—grid supply.

The proposed EMS architecture is designed and validated in Simulink and simple relational operators follow it. Relational operators can compare two constant values and determine whether they are true or false. The relational operator is straightforward, simple to understand and simple to use. The proposed approach focuses primarily on two different goals: (i) electricity cost reduction and (ii) carbon emission reduction (CO₂). The flowchart of the proposed method is shown in Figure 11.

$${\mathrm{P}}_{\mathrm{o}}={\mathrm{N}}_{\left(\mathrm{a}\mathrm{p}\mathrm{p}\right)}\times {\mathrm{A}\mathrm{p}\mathrm{p}}_{\left(\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{g}\right)}$$

(12)

$${\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}}_{\left(\mathrm{c}\mathrm{o}\mathrm{n}\right)}={\mathrm{P}}_{\mathrm{o}}\times {\mathrm{T}}_{\left(\mathrm{o}\mathrm{p}\mathrm{e}\mathrm{r}\right)}$$

(13)

$${\mathrm{C}\mathrm{o}\mathrm{s}\mathrm{t}}_{\left(\mathrm{d}\mathrm{a}\mathrm{y}\right)}={\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}}_{\left(\mathrm{c}\mathrm{o}\mathrm{n}\right)}\times {\mathrm{P}}_{\left(\mathrm{T}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{f}\mathrm{f}\right)}$$

(14)

where ${\mathrm{P}}_{\mathrm{o}}$—power output, ${\mathrm{N}}_{\left(\mathrm{a}\mathrm{p}\mathrm{p}\right)}$—number of home appliances, ${\mathrm{A}\mathrm{p}\mathrm{p}}_{\left(\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{g}\right)}$—power rating of home appliances, ${\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}}_{\left(\mathrm{c}\mathrm{o}\mathrm{n}\right)}$—total energy, ${\mathrm{T}}_{\left(\mathrm{o}\mathrm{p}\mathrm{e}\mathrm{r}\right)}$—operating time of home appliances and ${\mathrm{P}}_{\left(\mathrm{T}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{f}\mathrm{f}\right)}$—electricity tariff.

A sensor continuously monitors the solar-connected appliances. If the solar panel output energy is insufficient to power the devices, an error signal is generated by the instant sensors and sent to the controller. The load can acquire power from the grid based on the reference signal.

5.2. Fuzzy-Inference-System-Based Source Switching

In the proposed approach, source switching between grid and photovoltaic occurs in the HEMS by implementing a fuzzy logic controller using the Mamdani interface system. The total load is classified into two types: priority load 1, supplied by PV and priority load 2, supplied by the grid, as shown in

Table 7. Classification of loads.

Priority 2 Load		Priority 1 Load
Appliances	No. of Appliances	Appliances	No. of Appliances
Light	1	Light	3
Mixer/grinder	1	Fan	2
Grinder	1	TV	1
Iron box	1	Water purifier	1
Motor	1
Water heater	1
Refrigerator	1
Washing machine	1
Motor	1

. The fuzzy model of the proposed system is modeled by considering solar as input and grid as output with centroid defuzzification technique and maximum aggregation.

The membership functions of the input and output are shown in Figure 12 and Figure 13. The rules for the proposed condition are displayed in the rule editor, as shown in Figure 14. The setting of membership functions plays a significant role in the fuzzy inference system (FIS). In the proposed FIS, a triangular membership function has been used. The input solar membership plot ranges from 0 to 600 and has four membership variables, namely solar absent (0, 0, 0), less solar power (0, 225, 450), adequate solar power (450, 455, 460) and excess solar power (460, 550, 600). Similarly, the output grid membership plot ranges from 0 to 1 and has four membership functions, namely source (0, 0.1, 0.2), supply remaining load (0.2, 0.3, 0.5), supply excess power (0.5, 0.6, 0.7) and received energy (0.7, 0.8, 1).

The rules for the proposed system are as follows:

• When solar energy is absent, the grid feeds supply to both priority loads.
• When adequate solar energy is available, the priority 2 load will be handled by the grid supply.
• During the condition of power shutdown (the grid is absent), the priority 1 load will be fed by the grid.
• During reduced solar energy, grid energy manages the remaining power.
• During the production of excess solar power, excess power must be fed to the grid.

The output of the proposed rules is examined in the rule viewer. Figure 15a shows that when solar energy is zero, the grid supplies the load, as per rule 1. Figure 15b shows that when adequate solar energy is available, the grid supplies priority 2 loads, as per rule 2. Figure 15c shows that excess solar energy is supplied to the grid. The main drawback of FIS in the proposed objective is that implementation is complex when both solar and grid are considered as input and loads are considered output. The limit setting of the membership function needs more development to achieve high accuracy.

5.3. Artificial Neural Network-Based Source Switching

To overcome the difficulties of the fuzzy logic controller, artificial-neural-network-based source switching is proposed. An ANN is a computer model that tackles complicated problems in various applications by simulating the operation of a biological network of neurons. The proposed network has two inputs, namely solar energy and grid supply. The load is considered as output. Using MATLAB R2022a software, two different ANN networks with neuron counts of 6 and 10 were built and evaluated. The input of the ANN network uses a total of 46 samples that were collected from previously conducted experiments. The total data used are 40% for training, 30% for validation and 30% for testing. The scaled conjugate gradient (trainscg) algorithm function was utilized in the networks. After numerous trials and errors, the concealed layer’s number was finally decided upon in this investigation to be 10. A tangent sigmoid function was utilized for the hidden layer and for the output layer, a linear transfer function.

Table 8. ANN parameters considered for simulation.

Activation function—hidden layer	Tangent sigmoid function
Activation function—output layer	Linear transfer function
GPU type	No GPU used
Learning rate	0.1
Optimization algorithm	Scaled conjugate gradient (trainscg) algorithm

shows the training parameters considered for simulation.

The mapping of the output statement corresponding to the input is shown in

Table 9. Neural network output mapping.

Output	Statement	Load
000	Absence of solar energy and grid supply	No load is supplied
001	Presence of adequate solar and lack of a grid supply	Solar will supply priority 1 load
100	Lack of solar energy and presence of a grid supply	The grid will supply both priority 1 and 2 loads
010	Presence of adequate solar and grid supply	Solar will supply the priority 1 load; the grid will supply the priority 2 load
110	Excess solar supply	Solar will supply priority 1 load, the grid will supply excess 2 loads and the extra solar energy will be transferred to the grid

.

Table 10. Training performance of NN.

Neural Network with 10 Neurons		Neural Network with 6 Neurons
Epoch	10	Epoch	20
Gradient	0.0138	Gradient	0.0137
Validation check	6	Validation check	6
Training error	0.222	Training error	0.175
Validation error	0.100	Validation error	0.1364
Testing error	0.0500	Testing error	0.2500

shows the training parameters of both networks. Figure 16 depicts the variation in the gradient error and the number of validation tests at each network epoch with six neurons. According to this figure, the gradient error is 0.0137 and there are six validation checks at 20 epochs. Similarly, Figure 16 depicts the variation in the gradient error and the number of validation tests at each network epoch with ten neurons.

By analyzing the performance of both networks using the performance plots shown in Figure 17 and Figure 18, it can be seen that a neural network with 6 neurons has a cross-entropy error of 0.05 at the 14th epoch and a neural network with 10 neurons has a cross-entropy error of 0.1 at the 4th epoch, as shown in Figure 19.

An NN with 10 neurons is selected for simulation and tested for its performance. The input of 200 W solar energy and 4500 W grid supply is passed to the Simulink model and the result obtained as output is 100, as shown in Figure 20, which indicates the absence of solar energy and that the supply of both priority 1 and 2 loads is provided by the grid supply. By applying source switching, the demand on the grid is reduced by supplying the load with solar power. Excess solar energy can also be fed to the grid, so that the customers receive an incentive from the government and the cost of electricity is also reduced. The performance of the system is shown in

Table 11. Overall system performance.

Bill Month	Unit Consumed (kWh)	Bill Amount (INR)	${\mathbf{C}\mathbf{O}}_2$ Emission (Kg)	Unit Consumed (kWh)	Bill Amount (INR)	${\mathbf{C}\mathbf{O}}_2$ Emission (Kg)
	Conventional			Proposed Method
January–February	696	4074	592	539	2772	458
March–April	816	5170	694	659	3741	560
May–June	896	5970	762	739	4661	628
July–August	815	4700	693	658	3732	559
September–October	798	4992	678	641	3579	545
November–December	788	4442	670	631	3489	536
January–February	696	4074	592	539	2772	458

.

Table 11 shows that when a house’s total load is connected to the grid supply, it consumes 4809 kWh of energy, which costs INR 29,348 annually. Table 11 shows a high-priority load supplied by solar of about 3867 kWh of energy, costing INR 21,974 yearly. Hence, the demand on the grid and the electricity cost of the consumer are reduced simultaneously. The proposed HEMS is fully reliant on appliance scheduling; through this process, consumers’ quality of experience is disturbed. To avoid this discomfort, the proposed method implemented different priorities of appliances, as shown in Table 8. Due to this process, appliance scheduling is straightforward and organized.

The proposed method has the potential to significantly reduce ${\mathrm{C}\mathrm{O}}_2$ emissions from residential buildings. The ${\mathrm{C}\mathrm{O}}_2$ emission factor for the proposed system is calculated using the expression shown in Equation (15), while

Table 12. Electricity tariff description.

Domestic Consumers Slab for Bi-Monthly	Slab Categories Based on Energy Consumption (kWh)	Units (kWh)	Unit Charges (INR/kWh)	Fixed Charges Bimonthly (INR)
Energy Consumption Slab-1	Less than 500	1–200	2.25	30
Energy Consumption Slab-2		201–400	4.5	30
Energy Consumption Slab-3		401–500	6	30
Energy Consumption Slab-1	Above 500	1–200	4.5	30
Energy Consumption Slab-2		401–500	6	30
Energy Consumption Slab-3		501–600	8	30
Energy Consumption Slab-4		601–800	9	30
Energy Consumption Slab-5		801–1000	10	30
Energy Consumption Slab-6		1001 above	11	30

displays the energy consumption pattern and ${\mathrm{C}\mathrm{O}}_2$ emission factor for a single home over a year.

$${\mathrm{C}\mathrm{O}}_2=\mathrm{E}\mathrm{F}\times \mathrm{k}\mathrm{W}\mathrm{h}/\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}$$

(15)

where EF is the emission factor.

The source-switching approach is the main component of the suggested technique.

This method will involve switching transients and the household appliances will feature a mix of non-electronic and electronic devices. A non-intrusive load monitoring system observes each appliance in the house. All home appliances function together to create a complicated current waveform. A complex current is made by adding up all of the current harmonics. It is typically used to increase the power factor and decrease switching transients by injecting current into the switching process to lessen switching transients. For a better understanding of the switching transient in various household appliances, see Figure 21, which displays the current versus time curve. Due to the mix of electronic and non-electronic appliances in household loads, home appliances behave differently. The switching process is seamless and the appliances’ operation will not be disturbed.

• Electricity price is related to the weather conditions, global economic conditions, utility economic conditions, government policy, generation, transmission and distribution; these factors play major roles in electricity tariffs. In India, the cost of supply increases by an average of 8.3% per year [39]. The state government will provide a tariff increase in accordance with its energy and economic policies. The utility sector will follow the two different slab rate methodologies, which are less than 500 kWh and above 500 kWh, as shown on Table 12.
• The solar panel’s efficiency declines over time due to environmental factors such as weather, dust, material quality and power outages. According to these characteristics, solar panel performance decreases by 0.8% to 0.9% on average per year [40].
• Calculating the payback period requires taking into consideration increasing home electricity demand. In order to calculate the payback, the domestic consumer’s rising demand, which is estimated to be 8.01% between 2013 and 2019, was considered [41].

The payback period is the most important factor for consumers and is also the base for renewable power opportunities in India. This process encourages a larger number of consumers to implement solar installation in their residences. The following parameters influence the total payback period. The payback details of the proposed methods are shown in

Table 13. Payback period details.

Description	Cost (INR)
Cost of 1KW rooftop solar system	50,000
Subsidy at 40%	18,769
Cost after subsidy	28,154
Total electricity generation from solar plant (Average solar irradiation in Tamil Nadu, India, is 1266.52 W/sq.m, considering 5.5 sunshine hours)	1500 kWh
Financial consideration (Rs. 7.5/kWh—average)	11,250
Carbon dioxide emissions mitigated/year	1275 Kg
Payback period (months)	30

.

6. Conclusions

This study describes the development and application of an HEMS with a solar PV system to reduce ${\mathrm{C}\mathrm{O}}_2$ emissions and energy costs for communal dwellings. A proposal for ANN-based MPPT tracking has been made to maximize solar output power. The proposed energy management system is compared and contrasted three techniques, the relational operator approach, the fuzzy logic controller and the artificial neural network, to minimize overall energy demand, cost and emissions without compromising customer satisfaction. Neural networks are the most effective solution since they require less computing time than relational operators and are more challenging to frame as fuzzy logic controllers. The consumer’s electricity bill and emissions are decreased when solar energy is used at home to power high-priority loads. The neural model achieves the target with a reduced RMSE error of less than 0.1%. The results show that an ANN-based HEMS system might be preferred for residential appliances as it does not impair consumer experience. In our planned work, priority loads are categorized and an ANN for each load will be implemented based on operating power and time. Applying energy from renewable sources for residential loads reduces the stress on the grid, increases customer satisfaction, reduces electricity cost and reduces ${\mathrm{C}\mathrm{O}}_2$ emission.