An integrated framework of Bi-directional long-short term memory (BiLSTM) based on sine cosine algorithm for hourly solar radiation forecasting

Abstract Accurate and reliable solar radiation forecasting is of great significance for the management and utilization of solar energy. This study proposes a deep learning model based on Bi-directional long short-term memory (BiLSTM), sine cosine algorithm (SCA) and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) for solar radiation forecasting. Firstly, the CEEMDAN is applied to decompose the stochastic historical time series into certain periodic intrinsic mode functions (IMFs) and a residual. Secondly, the significant antecedent solar radiation patterns of the decomposed sub-modes are identified via two statistical techniques, namely, the autocorrelation function (ACF) and the partial autocorrelation function (PACF). Thirdly, all the sub-modes are forecasted using the BiLSTM model, and the parameters of the BiLSTM model are optimized using the SCA algorithm. Finally, the forecasted sub-modes are aggregated to generate the final forecasting result. The accuracy of the proposed deep learning model is investigated by applying it in forecasting hourly solar radiation of four real-world datasets over multi-step horizons. Comparative experiments with other seven models demonstrate the effectiveness of the integrated model, the CEEMDAN technique and the SCA algorithm, respectively. The proposed model can obtain higher prediction accuracy than the existing models for all datasets and forecasting horizons.