Planning in the operation and expansion of electrical systems must consider these changes and one of the current and practical tools to improve the technical and economic operating conditions of power systems is reactive power compensation.
1.1. Literature Review
Electric power systems have become increasingly large and complex, and their operation in terms of voltage stability is at its limit with a small safety margin, which is why this research’s literature review shows different classical optimization techniques applied to reactive power compensation and new techniques based on autonomous learning that are applied to solve the problems that arise in distribution and transmission networks (especially focusing on voltage profiles improvement and losses reduction).
In [
3], a detailed analysis of the different objectives that are pursued in the distribution and transmission systems regarding sizing and optimal placement of reactive power compensation devices is shown. This research also indicates existing technologies related to FACTS as well as the different optimization criteria to improve the operating and economic conditions of the EPS.
In [
1], new techniques that have been developed in machine learning are shown, the most important techniques are ANN (artificial neural networks) in combination with control techniques such as fuzzy control. This research focuses on stability improvement by locating reactive compensation in the electrical power system.
The research in [
4] analyzes the location and sizing of FACTS compensators, specifically D-STATCOMP, using a hybrid heuristic combining the algorithm based on the prey targeting behaviour of whales and grey wolf optimization. The methodology is tested in distribution systems for power quality improvement and voltage stability and its performance is compared with conventional optimization methods GA, ABC, PSO, GWO and WOA to determine its effectiveness.
In [
5], the authors use particle swarm optimization (PSO) to determine the capacity and location of capacitor banks to guarantee the operation of a microgrid under minimum cost criteria, improving the power factor and reducing the maximum and average voltage deviations of the system.
BCC capacitor banks for reactive compensation in distribution systems that are used to improve power factor, power losses reduction and minimization of annual operating cost are discussed in [
2]. This research proposes a hybrid method based on particle swarm optimization together with a particle swarm gravitational search algorithm (PSOGSA) as an optimization mechanism to solve the problem of optimal BCC allocation with minimization of annual operating cost and improvement of system power quality and propose a voltage-loss cost index (VLCI) as an allocation mechanism. The methodology is tested in the 33- and 69-bus radial systems, as well as on a real distribution network of 111 nodes in Moscow.
In [
6], the location and sizing of the UPQC power quality conditioner are proposed, considering the reconfiguration of the distribution network for loss reduction all under an exhaustive search. The model is tested in a radial-type network of 69 nodes.
In [
7], a convex-genetic method is proposed to determine the location and size of a D-STATCOM to reduce power losses and minimize the annual operating costs of a 33-node radial distribution network. The results show a processing time for the algorithm of 3.11 h.
The investigation in [
8] shows a transmission system planning strategy that considers the optimization of several costs, among them, active power losses, reactive power generation, FACTS devices and transmission line loadability, through a probabilistic hybridization of the crow search algorithm and JAYA. The heuristic is validated in two test systems of 30 and 75 nodes.
In [
9], the research uses the dolphin algorithm to determine the location and size of reactive compensation in 16- and 33-busbar distribution networks under different models for various types of loads (constant and mixed power, current and impedance), minimizing power losses and voltage drops between nodes.
In [
10], a methodology is presented to optimize the cost/benefit ratio in the installation of FACTS in distribution networks employing a differential evolution algorithm considering the lowest capacity and cost of the FACT device and minimizing the power losses in the network. The methodology is evaluated in systems of 30, 33 and 69 nodes.
In [
11], the research proposes a heuristic that considers the modified artificial bee colony algorithm for optimal reactive compensation placement through UPQC to improve the power quality of the IEEE 30-node network and compares the results with other methods such as PSO and the genetic algorithm.
In [
12,
13], a planning methodology is established for the simultaneous location of reactive compensation devices, specifically the OUPQC with its series and shunt units (SEU and SHU) using the Cuckoo search algorithm and load growth. The methodology is tested in the 69-node IEEE network in steady state and the results are effective in improving the quality of the network in economic and technical terms.
In [
14], the location of reactive compensation in an electrical power system is established and analyzed by employing deep neural networks to improve the voltage profile in each of the nodes of the evaluated systems (14, 30 and 118 nodes). The standard deviation of all voltage profiles over a referential value is evaluated to determine the power flows and training data of the neural network and finally to determine the optimal location of the reactive compensation.
In [
15], an application based on fuzzy logic and bio-genetic algorithms is established to help decision-making in the field of operation and reactive power compensation of an islanded power grid. The research’s objective is to improve power quality in terms of power factor, voltages and loss reduction.
In [
16], the planning of transmission networks to improve long-term operating conditions is analyzed by using probabilistically modelling of load scenarios as well as capital and operating costs in the face of massive installation of FACTS devices to minimize generation cost. The heuristic algorithm consists of a sequence of quadratic schedules solved by CPLEX and tested on the IEEE 30-bus system and extended to a 2736-bus model in Poland.
In [
17], a methodology is established to compensate for the reactive power by minimizing the total harmonic distortion of voltage and current as well as the installation cost of the passive filter. The optimization problem is solved with a learning algorithm and Pareto distribution.
The work in [
18] presents a methodology for optimal reactive power compensation in electrical microgrids using a multi-criteria decision algorithm based on heuristic methods.
Research in [
19] presents a model to solve capacitor allocation considering PV penetration and feeder reconfiguration simultaneously in a distribution network minimizing capacitor costs and power losses. The model uses PSO to solve the allocation problem and is tested on a 33-node IEEE network.
In [
20], a heuristic based on the whale optimization algorithm is proposed for the optimal placement, sizing and coordination of FACTS devices, specifically TCSC, SVC and UPFC. In the transmission network, the management of the algorithm allows for minimizing operating costs, energy losses and voltage deviations, and is further compared with genetic algorithm (GA) and particle swarm optimization (PSO). The research results show that the total system operating costs and transmission line losses were significantly reduced with the WOA method compared with existing meta-heuristic optimization techniques.
In [
21], the research compares a probability-based heuristic with an artificial neural network for optimal sizing and location of distributed generation arrays and flexible AC transmission systems to improve DG-FACTS for operation improvement of two 9-bus and 57-bus systems. The results show that the neural network is more efficient in terms of DG-FACTS capacity sizing compared to the heuristic.
In [
22], three different evolutionary techniques, the big bang–big crunch algorithm, gravitational search algorithm and bacterial foraging algorithm are used in conjunction to train and validate an ANN. This research’s strategy manages reactive power and corrects the voltage profile. Additionally, as a study case, the IEEE 30 busbar system was used. The results showed an improvement in the system’s voltage profiles.
As has been stated in this section, reactive compensation has been widely studied and optimal location and sizing have been analyzed with optimization techniques, fuzzy logic and probabilistic logic, among others. However, all those techniques need considerable processing power and their processing time is quite considerable as well. A summary of the main strategies used in the literature review for reactive compensation through optimization algorithms or artificial neural networks is shown in
Table 1.
Therefore, this research proposes a methodology which aims to improve the voltage profile in the whole transmission system under scenarios in which a PQ load is randomly connected to any busbar of the system. The optimal location of sizing of reactive compensation will be found through deep neural networks, which require significantly less processing power and are faster than optimization algorithms described in the literature review.