Solar Photovoltaic Power Generation Prediction based on Deep Learning Methods

Abstract

In recent years, renewable energy power generation has received more and more attention. Since the forecast of electricity generation is helpful for properly using and managing electricity. Therefore, this study uses time series analysis and deep learning methods, Long Short-Term Memory (LSTM), Temporal Convolutional Network (TCN), and Gated Recurrent Unit (GRU), to forecast solar power generation. Furthermore, this study also uses different time intervals (every ten minutes, every thirty minutes, hourly, daily) to forecast the power generation and evaluate their performances. In comparing the four deep learning models, the prediction performance of LSTM is the best, while the performance of the TCN model is poor. In addition, the time interval length greatly influences the prediction performance. The time interval is divided into smaller, and the performance of various deep learning models is relatively good and stable; otherwise, the performance of the models is poor.