Integrating WRF Regional Climate Model with Neural Network Models for Spatio-Temporal Photovoltaic Power Output Forecasting

This paper endeavours to contribute to the improvement of short and medium term forecasting of photovoltaic (PV) power output. The study is based on a statistical analysis and a forecasting model of PV power output that uses meteorological parameters. Given the complex and marked topography of Reunion Island, the approach adopted here is to generate meteorological data over the whole island at a high resolution and over several years using the regional climate model Weather Research Forecasting (WRF). A statistical forecasting model, based on the Johansen cointegration method, is proposed using the correlation between selected meteorological variables and PV power production. This model allows to estimate the PV power output for the whole territory of Reunion island. In this paper, a spatio-temporal model using long short-term memory (LSTM) is developed in order to produce efficient forecasts over the whole island at three time horizons likely to be of interest to a network manager: M+1, d+1 and h+1