AI-based solar energy forecasting for smart grid integration

Abstract

Integrating solar energy power into the existing grid system is a challenging task due to the volatile and intermittent nature of this power. Robust energy forecasting has been considered a reliable solution to the mentioned problem. Since the first success of Deep Learning models, it has been more and more employed for solving problems related to time series forecasting and excellent results were achieved. In this work, we propose a hybrid method based on the combination of an LSTM neural network and an autoencoder. The LSTM neural network extracted temporal features of the historical time series of solar energy production. The spatial features were extracted by the autoencoder, and predictions have been generated. To demonstrate the robustness of the proposed approach, four error metrics were used. The proposed method was compared to the state-of-the-art models, and superior results were achieved. A low SDE of 0.62 and an RMSE of 0.60 were achieved. Experimental results have proved the efficiency of the proposed approach and proved that extracting the temporal features before the spatial features is better than extracting spatial features than extracting temporal features.