Forecasting Photovoltaic Power Output Using Long Short-Term Memory and Neural Network Models

Abstract

The unpredictable and stochastic nature of solar energy brings forth an array of challenges to the planning, management, and operation of power grid systems as the fluctuation in the output can lead to cost increases, difficulty in grid integration and also cause issues with control and reliability of the system. Hence, forecasting of photovoltaic (PV) output assumes greater significance as it helps operators manage changes in the output and organize optimal schedules for power generation. This paper presents two deep learning models, Long Short- Term Memory and Back Propagation Neural Network, for forecasting PV power output and the comparison of their MSE values for the annual period. The input data was refined initially by performing correlation tests and accordingly wind speed was eliminated from the input dataset. The optimal MSE values for LSTM and BPNN network were 0.000626 and 0.1547 respectively. Both the models preformed effectively and LSTM model performed better than BPNN model due to better generalization capability. These modeling approaches can be employed for forecasting the future solar power output of a PV system to assist in optimal scheduling and planning of power grid system.