Wind speed forecasting in high-speed railway system based on secondary decomposition and gated recurrent unit network

Accurately predicting wind speed (WS) is crucial for the operational safety of high-speed railway (HSR) systems in strong wind environments. However, the prediction task is challenging due to WS data's nonlinearity and intermittency nature. To address this issue, this study proposes a hybrid prediction model SD-BGRU, which combines secondary decomposition (SD), sample entropy (SE), and gated recurrent unit (GRU). Specifically, the raw WS signal is first decomposed using complete ensemble empirical mode decomposition adaptive noise (CEEMDAN). The decomposed data are clustered into three groups based on their SE. VMD is then applied to further decompose sub-signals with high SE values. The remaining sub-signals obtained from CEEMDAN and sub-signals obtained from VMD are fed into bidirectional GRU (BGRU) to generate individual predictions. These prediction results are combined to produce a final prediction result. We conducted extensive experiments on a real-world WS dataset to evaluate the performance of SD-BGRU. The experiment results demonstrate that SD-BGRU outperforms other comparative models, with a mean absolute error of 0.3395, root mean square error of 0.4571, and mean absolute percentage error of 8.88%. These promising results indicate that SD-BGRU is an effective way for guiding decision-making in the HSR system.