A Comparative Study of Machine Learning–based Models for Short-Term Multi-step Forecasting of Solar Power: An Application for Nhi Ha Solar Farm

Abstract

Over the past few decades, the utilization of solar power has gained immense significance in the power grid, gradually taking over the responsibilities of fossil fuel-based power. Therefore, accurate short-term forecasting of photovoltaic power output is crucial for making informed decisions regarding power generation, transmission, and distribution. Consequently, many machine-learning models were used to reliably forecast solar power. In this study, four machine learning models have been studied which are Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Long short-term memory (LSTM) and Extreme learning machine (ELM). They have been used to forecast the solar power of Nhi Ha solar farm in short-term. First, data from Nhi Ha solar farm were collected and underwent preprocessing before being utilized by distinct machine learning models. The Root Mean Squared Error (RMSE) and normalized RMSE (N-RMSE) obtained from the models will be analyzed to determine the most effective model for short-term solar power forecasting. Following a comprehensive analysis, it has been determined that all four models have produced favorable outcomes, with low values of RMSE and N-RMSE indicating high levels of reliability and accuracy. Of the models considered, the LSTM and ELM models have demonstrated superior performance, making them the optimal choice for precise and dependable short-term solar power forecasting.