Sensibility Analysis with Genetic Algorithm to Allocate Distributed Generation and Capacitor Banks in Unbalanced Distribution Systems

Abstract

In the last years, many authors have dedicated themselves to solving the problem of distributed resources allocation in Distribution Systems (DS). The optimal siting and sizing of such equipment can improve the efficiency of the provided service, promoting technical and economic gains. Several papers use artificial intelligence algorithms and methods based on population evolutions to achieve this goal, due to the increasing complexity of the simulated systems and the highly combinatorial characteristics of this problem. Reducing the search space and/or setting the initial population are critical factors for improving the performance of population-based optimization methods. In the literature, the adjustment of the initial population does not receive the attention it should, especially when dealing with three-phase large-scale unbalanced systems – when computational time becomes primary adversity. The objective of this paper is to propose a novel approach for DS optimization, using Genetic Algorithms (GA) to map which areas are more sensitive to equipment allocation to reduce losses. In this work, capacitor banks and distributed photovoltaic (PV) generation are considered. The method adopted to carry out the allocation considers proper modeling of unbalanced DS through multiphase Optimal Power Flow frameworks. Its development takes advantage of the probabilistic nature of the GA to perform a sensitivity analysis to determine the positions that may reduce the losses in a daily planning horizon. For this, the execution of GA is done sequentially, under different conditions of generation and load, selected at random, setting in optimization a 24-hour load and PV generation curves that take into consideration the uncertainties, representing each hour of the day individually for better detail of the planning horizon. It was seen that the best positions are always concentrated in specific regions, called Optimal Allocation Zones (OAZ), which may favor the planning of Utilities. Furthermore, the OAZ method can be used to define an initial population of evolutionary algorithms, thus achieving a reduction in computational time and a better quality of solutions, which is a major contribution to literature. Through this approach, the computational time can be reduced up to 60% in large-scale systems. Simulations were carried out on four unbalanced distribution test-feeders: IEEE 4, IEEE 13, IEEE 37, and IEEE 123 Node Test Feeders.