Predicting renewable energy production outputs from climate factors: A machine learning approach

Abstract

Abstract Predicting the energy output of renewable energy systems is a growing field of research that goes in parallel with advances in machine learning (ML) methods. However, the complexity of those ML methods along with the variety of renewable energy sources used in prediction models requires the development of highly robust approaches. The automated ML framework proposed in this study streamlines the steps involved in model development including data processing, model construction, hyper-parameter optimization and inference deployment. This paper also identifies the factors affecting the performance of ML methods such as sampling, encoding categorical values, feature selection, and hyper-parameter optimization for different algorithms. This paper presents an automated ML approach for a variety of applications in the renewable energy domain. The proposed automated ML framework is used to compare a variety of methods combined with alternative training/test ratios.