Day-Ahead and Week-Ahead Solar PV Power Forecasting Using Deep Learning Neural Networks

This paper presents the development of a deep learning model using Long Short-Term Memory Neural Network (LSTMNN) to perform short-term forecasts for solar PV power generation for multiple seasons of the year. Other artificial neural network (ANN) models such as feed-forward Extreme Learning Machine (ELM) and the shallow learning Elman Neural Network (ENN) are compared with the proposed LSTMNN model to perform day-ahead and week-ahead PV power forecasts. The LSTMNN is the preferred tool for solving time-series forecasting problems as it can predict future values based on previous sequential data. All three machine learning models (LSTMNN, ELM, and ENN) are trained using historical meteorological data, i.e., Solar Radiation (SR), and temperature, as well as using solar PV power generation data. To validate the performance of the forecasting accuracy of the models, accuracy measures are calculated. Test results obtained from various cases for multiple seasons of the year demonstrate the higher degree of accuracy of the proposed LSTMNN model.