The large-scale integration of wind power faces persistent schedule-tracking challenges due to uncertainty and intermittency. To address this challenge, this study develops a predictive control framework for coupled wind–hydrogen systems that combines high-fidelity wind power forecasting with rolling-horizon, multi-objective scheduling. At its core is a “decompose–cluster–redecompose” forecasting pipeline that performs two-stage signal decomposition using improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and variational mode decomposition (VMD). The decomposed components are modeled using a convolutional neural network–bidirectional long short-term memory (CNN–BiLSTM) architecture, whose hyperparameters are tuned with the Tornado Optimizer with Coriolis Force (TOC), thereby delivering high-accuracy, short-term wind power forecasts. Building on these forecasts, this study formulates a multi-objective optimization model that (i) maximizes plan-tracking accuracy, (ii) minimizes joint output fluctuations, and (iii) preserves the health of the energy storage subsystem. The model is solved efficiently using a hybrid particle swarm–adaptive whale optimization algorithm (PSO–AWOA) in a rolling-horizon manner. Using real-world data, this study shows that the proposed strategy reduces the plan-tracking RMSE from 16.18% to 9.05%, lowers the average system output fluctuation from 6.5265 MW to 0.1933 MW, and decreases the mean predicted power deviation from 8.7630 MW to 0.6475 MW, compared with conventional control strategies. It also mitigates power shocks in the storage system and excessive adjustments, enabling cost-effective operation within safe limits. These results demonstrate the improved dispatchability and grid friendliness of large-scale wind power integration.
Dai et al. (Mon,) studied this question.