A Solar Irradiance Forecasting Framework Based on the CEE-WGAN-LSTM Model

Abstract

With the rapid development of solar energy plants in recent years, the accurate prediction of solar power generation has become an important and challenging problem in modern intelligent grid systems. To improve the forecasting accuracy of solar energy generation, an effective and robust decomposition-integration method for two-channel solar irradiance forecasting is proposed in this study, which uses complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), a Wasserstein generative adversarial network (WGAN), and a long short-term memory network (LSTM). The proposed method consists of three essential stages. First, the solar output signal is divided into several relatively simple subsequences using the CEEMDAN method, which has noticeable frequency differences. Second, high and low-frequency subsequences are predicted using the WGAN and LSTM models, respectively. Last, the predicted values of each component are integrated to obtain the final prediction results. The developed model uses data decomposition technology, together with advanced machine learning (ML) and deep learning (DL) models to identify the appropriate dependencies and network topology. The experiments show that compared with many traditional prediction methods and decomposition-integration models, the developed model can produce accurate solar output prediction results under different evaluation criteria. Compared to the suboptimal model, the MAEs, MAPEs, and RMSEs of the four seasons decreased by 3.51%, 6.11%, and 2.25%, respectively.

Keywords

• time-series forecasting
• deep learning
• signal decomposition
• solar irradiance