Análise de séries temporais utilizando Aprendizado de Máquina ao Extremo (ELM) para previsão de potência em painéis solares

Abstract

Thinking about the importance of renewable energies and their technological developments, the use of computational and statistical tools to assist in their administration and management, can be a next step towards growth and encouragement in the use of these sustainable energy sources, such as photovoltaic plants. Using time series and their analysis methods to describe the sample set, stochastic statistical forecasts can be made with them and by applying the intelligent model of Extreme Learning Machine (ELM) there is an improvement in the accuracy of these forecasts. Taking advantage of finding the best hyperparameters at ELM initialization, the use of optimization meta-heuristics such as: Coyote Optimization Algorithm(COA), Differential Evolution (DE) and Particle Swarm Optimization (PSO), refine the forecast results and it is also possible to make a comparison between them, indicating the best one for the application in question with the use of forecast error minimization as a comparison metric. With this, it is possible to notice promising results in an accuracy close to 98% with the PSO meta-heuristic.