Comparative Analysis of Machine Learning Models for Short-Term Net Load Forecasting in Renewable Integrated Microgrids

Abstract

Accurate net load forecasting (NLF) is crucial in modern power systems and microgrids to ensure optimal operation and management. At the microgrid level, the increasing penetration of renewable energy sources requires more efficient methodologies for NLF, since statistical approaches fail to provide accurate forecasts. Performance limitations of existing statistical approaches can be overcome by leveraging machine learning (ML) models. The purpose of this work is to compare the forecasting performance of six ML models (artificial neural network, extreme gradient boosting, k-nearest neighbors, random forest, recurrent neural network and support vector regression) and identify the best-performing model for short-term net load forecasting (STNLF). The comparative analysis was carried out using historical net load and weather data from the renewable integrated microgrid of the University of Cyprus. The results demonstrated accuracies below 10% for all STNLF ML models. Random forest was the best performing model, achieving a normalized root mean square error of 4.32%. The findings illustrate the applicability of STNLF ML models in renewable integrated microgrids, which can benefit microgrid operators in managing and controlling their various assets.