@Article{ee.2026.074643,
AUTHOR = {Mingxuan Ji, Jing Gao, Dantian Zhong, Yingqi Xu, Shuxiang Yang, Zhongxiao Du, Yingming Liu},
TITLE = {A Short-Term Wind Power Forecasting Method Based on Adaptive BKA-TCN-BiLSTM Hybrid Model with AP Clustering},
JOURNAL = {Energy Engineering},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/energy/online/detail/25635},
ISSN = {1546-0118},
ABSTRACT = {The intermittency of wind power poses severe challenges to the safe and stable operation of power grids, while conventional forecasting models are deficient in prediction accuracy and adaptability to variable weather conditions. To address these issues, this study proposes an adaptive short-term wind power forecasting model integrating affinity propagation (AP) clustering and a black-winged kite algorithm (BKA)-optimized temporal convolutional network-bidirectional long short-term memory (TCN-BiLSTM) hybrid architecture. First, mutual information was employed to screen key meteorological features, and AP clustering categorized historical data into six distinct weather scenarios. A scenario-specific TCN-BiLSTM model was then constructed for each cluster: TCN was utilized to capture multi-scale local temporal features, while BiLSTM modeled global sequence dependencies, with BKA implementing global hyperparameter optimization. Final predictions were generated by invoking the model corresponding to the nearest cluster center. Comparative experiments against baseline models (LSTM, BiLSTM, CNN-LSTM, TCN-BiLSTM) demonstrate that the proposed model achieves remarkable performance gains: normalized root mean square error (nRMSE), normalized mean absolute error (nMAE), and normalized mean square error (nMSE) are reduced by over 1.27%, 1.73%, and 12.09%, respectively, with the coefficient of determination (R<sup>2</sup>) reaching 0.9084. This verifies substantial improvements in prediction accuracy and data fitting capability. The scenario-based modeling framework combined with intelligent hyperparameter optimization effectively enhances the model’s adaptability to complex weather, confirming its high accuracy and strong generalization for wind power forecasting tasks.},
DOI = {10.32604/ee.2026.074643}
}