Optimized LSTM Neural Networks Model applied for Solar PV Power Prediction

Abstract

Energy and climate challenges have driven significant growth in solar power generation. However, solar power production is intermittent and unstable, which complicates its integration into power grids. Techniques of forecasting Photovoltaic (PV) energy are needed to ensure its security and cost-effectiveness. This paper deals with the use of artificial intelligence techniques to predict photovoltaic power. The proposed techniques are Weighted Linear Regression (WLR) and Long Short-Term Memory neural networks (LSTM). The investigated data in this study was obtained from a photovoltaic solar power plant located in a specific geographical area. The studied parameters are wind speed (WS), ambient temperature (T), relative humidity (RH), irradiance (GHI) and wind direction (WD). The accuracy and ability of the LSTM model to explain data variation were examined by comparing its predictions with those of the WLR model, based on three performance measures: RMSE, MAE and R2. The obtained results show that weighted linear regression produces an acceptable estimation of photovoltaic power. However, the LSTM model performs better and has the potential to improve forecast reliability and accuracy.