Evaluación de un algoritmo basado en Machine Learning para un flujo de potencia óptima de corriente alterna ACOPF

Abstract

In this work, the feasibility of using machine learning (ML) to obtain solutions to the alternating current optimal power flow problem ACOPF (Alternating Current Optimal Power Flow) is analyzed. Because ACOPF is a nonconvex problem with high nonlinearity, numerous efforts have been made to find efficient optimization methods that can substantially reduce resolution times. OPF (Optimal Power Flow) problems are usually solved by interior point methods [1], also known as barrier methods. One of the most widely used approaches is the primary dual interior point technique with a filter line search [2]. These methods are robust but expensive, since they require the calculation of the second derivative of the Lagrangian in each iteration. A new and fruitful research direction is to use ML techniques to solve problems of operation and control of electrical networks. ML has been shown to significantly reduce the use of computational resources in many real-world problems. Several solution methods have been used, among them random forest, multi-objective decision tree and extreme learning machine [3, 4]. The ML operation in this case is applied as a method that first predicts voltage magnitudes and angles on each bus. Using network equations based on physics to calculate the injection of power in different buses. For general ML learning, the data is divided into three sets: one for training, one for validation, and finally, one for testing. These algorithms focus on minimizing their objective function and the cost of operating an AC transmission network.