Abstract
High-precision wind power forecasting is essential for grid scheduling and renewable energy utilization. Wind data's nonlinear, stochastic, and multi-scale characteristics create prediction challenges. This study proposes a hybrid model integrating adaptive improved singular spectrum analysis (ISSA), optimized bidirectional temporal convolutional network-bidirectional long short-term memory (BiTCN-BiLSTM) networks, and AdaBoost ensemble learning. Adaptive ISSA provides parameter-free, data-driven modal decomposition to reduce noise. Hybrid strategy-enhanced dung beetle optimization (OTDBO) fine-tunes hyperparameters of BiTCN-BiLSTM, and AdaBoost dynamically corrects errors, significantly improving robustness. Tests using seasonal datasets from Dabancheng wind farm (China) show substantial performance improvement (mean absolute error [MAE] reduced by 45.4%, root-mean-square error (RMSE) by 47.6%, p < 0.001), and training time reduced by 12.1%-21.3%. This method offers accurate, scalable forecasting for reliable renewable energy integration.