LSTM and Evans model for next 24 hours solar photovoltaic power output forecasting.

Abstract

This paper presents the results of forecasting the power output of a photovoltaic array for the next 24 hours from irradiation and temperature data by combining a Long ShortTerm Memory (LSTM) neural networkand the Evans model. The LSTM network allowed the forecast of irradiation and temperature as time series. Then, thanks to the forecasted values, the Evans model was used to estimate the photovoltaic power for each pair of predicted values from the panel parameters. The data used were obtained from the NASA databank for a period from 1st January 2010 to 31st December 2021 for Benin City Abomey-Calavi. The forecasted data were then validated experimentally by comparing them to the measured values in the field. For the experimental validation, for irradiation, temperature and PV power respectively Mean Square Error (MSE) values of 4.9 W/m2, 1.2 °C, 1.76 W were obtained and values of 0.98, 0.96, 0.97 were obtained for the regression coefficient (R2).