Deep Learning-Based Multistep Forecasting for Photovoltaic Power Generation: A Comprehensive Comparative Study

Electrical energy being the major contributor to the industrial revolution that has impacted social and economic growth has become an essential part of human life. The shift from finite, non-renewable power sources to sustainable, renewable options has been a significant consciousness within the power quarter. Solar energy is the most effective and common renewable energy resource. This study proposes an effective model to predict solar energy 24 hour ahead based on weather forecasts for successful integration into solar grids. Deep learning algorithms such as LSTM, Bidirectional LSTM BI-LSTM and GRU are implemented for energy prediction. To measure the model performance Quadratic Mean Error (QME i.e., MSE), Mean absolute deviation (MAD ie., MAE), Quadratic mean squared error (QMSE ie., RMSE) and R2-Score. Prediction using LSTM, BI-LSTM and GRU yielded R2-score error of 0.932, 0.930 & 0.919 respectively. Based on the results the LSTM provided a robust model for predicting 24 hr. ahead solar power.