Dynamic Fusion of Dual-Branch KAN-Enhanced Networks for Wind Power Forecasting via Cross-Modal Deep Reinforcement Learning

ZHANG Xin; FU Wenlong; LIU Yuan; KONG Zehao; LI Nan; CHEN Xi

PDF(3367 KB)

Electric Power Construction ›› 0

Dynamic Fusion of Dual-Branch KAN-Enhanced Networks for Wind Power Forecasting via Cross-Modal Deep Reinforcement Learning

ZHANG Xin¹, FU Wenlong^1,2, LIU Yuan¹, KONG Zehao¹, LI Nan¹, CHEN Xi¹

Author information +

History +

Abstract

[Objective] Affected by meteorological variations and the non-stationary characteristics of wind power generation,wind power sequences exhibit strong stochastic fluctuations,making it difficult for traditional prediction models to achieve effective modeling. To improve the accuracy of short-term wind power forecasting,this paper proposes a forecasting method based on a dual-branch KAN-enhanced model and cross-modal deep reinforcement learning. [Methods] First,meteorological features are selected using the Pearson correlation coefficient. Subsequently,an improved beaver behavior optimization (IBBO) algorithm is employed to optimize the parameters of variational mode decomposition (VMD),thereby achieving multi-scale decomposition of wind power sequences. Furthermore,a dual-branch prediction framework is constructed,where the KAN-enhanced Transformer model is utilized to extract global temporal dependencies,while the TCN-KAN model is designed to capture local dynamic patterns. On this basis,a cross-modal attention-driven proximal policy optimization (PPO)-based dynamic fusion module is introduced to adaptively integrate the prediction results from multiple models,ultimately obtaining accurate wind power forecasting results. [Results] Experiments are conducted using the operational data collected from a 99 MW wind farm throughout the year 2020,and comparisons are made with several baseline forecasting models. The results demonstrate that the proposed method achieves superior forecasting accuracy on the independent test dataset covering the entire year. [Conclusions] The proposed method effectively improves short-term wind power forecasting performance and exhibits good predictive stability under year-round operational conditions.

Key words

wind power forecasting / improved beaver behavior optimization algorithm / dual-branch forecasting model / proximal policy optimization / dynamic fusion

Cite this article

EndNote

Ris (Procite)

Bibtex

Download Citations

ZHANG Xin, FU Wenlong, LIU Yuan, KONG Zehao, LI Nan, CHEN Xi. Dynamic Fusion of Dual-Branch KAN-Enhanced Networks for Wind Power Forecasting via Cross-Modal Deep Reinforcement Learning[J]. Electric Power Construction. 0

References

[1] 朱嘉宁, 王钊, 李奕陶, 等. 基于时空图谱聚类与时序二维化Transformer的风电集群功率多尺度融合预测方法[J]. 电力建设, 2026, 47(3): 39-50.
Zhu Jianing, Wang Zhao, Li Yitao, et al.Multi-scale fusion prediction method for wind power cluster power based on spatio-temporal graph spectral clustering and time-series 2D-reshaping transformer[J]. Electric Power Construction, 2026, 47(3): 39-50.
[2] 张丁予, 解佗, 马易晨, 等. 基于爬坡特征识别的短期风电功率集成预测方法[J]. 电网与清洁能源, 2024, 40(8): 128-133.
Zhang Dingyu, Xie Tuo, Ma Yichen, et al.Research on integrated short term wind power prediction methods based on climbing feature recognition[J]. Power System and Clean Energy, 2024, 40(8): 128-133.
[3] 王凌梓, 沈海波, 邓力源, 等. 基于相似日聚类与WOA-BiLSTM-Copula算法的短期风光功率相关性概率区间预测方法[J]. 南方电网技术, 2025, 19(8): 44-52.
Wang Lingzi, Shen Haibo, Deng Liyuan, et al.Short-term wind and photovoltaic power correlation probability interval prediction method based on similar day clustering and WOA-BiLSTM-copula algorithm[J]. Southern Power System Technology, 2025, 19(8): 44-52.
[4] 傅质馨, 唐鹏, 王健, 等. 融合多元台风气象因素的海上风电功率预测[J]. 智慧电力, 2026, 54(5): 89-97.
Fu Zhixin, Tang Peng, Wang Jian, et al.Offshore wind power prediction incorporating multivariate typhoon meteorological factors[J]. Smart Power, 2026, 54(5): 89-97.
[5] 王小明, 徐斌, 尹元亚, 等. 基于空间相关性与Stacking集成学习的风电功率预测方法[J]. 电力工程技术, 2024, 43(5): 224-232.
Wang Xiaoming, Xu Bin, Yin Yuanya, et al.Wind power prediction method based on spatial correlation and Stacking ensemble learning[J]. Electric Power Engineering Technology, 2024, 43(5): 224-232.
[6] 张涛, 郝正航, 徐玉韬, 等. 基于多智能体强化学习的两阶段电压控制策略[J]. 南方电网技术, 2024, 18(12): 77-86.
Zhang Tao, Hao Zhenghang, Xu Yutao, et al.Two-stage voltage control strategy based on multi-agent reinforcement learning[J]. Southern Power System Technology, 2024, 18(12): 77-86.
[7] Wei J X, Zhang W Q, Zhang W J, et al.DBSTN: a dual-branch spatio-temporal network for wind power prediction using multi-modal fusion[J]. Energy, 2025, 341: 139471.
[8] Lu P, Ye L, Pei M, et al.Short-term wind power forecasting based on meteorological feature extraction and optimization strategy[J]. Renewable Energy, 2022, 184: 642-661.
[9] Dias F, Naik A.Hybrid wavelet Bi-LSTM model for stable short-term wind power forecasting[C]//2025 8th International Conference on Algorithms, Computing and Artificial Intelligence (ACAI). IEEE, 2025: 1-5.
[10] 司雅昕, 刘训强, 苏向敬, 等. 海上风电功率预测研究综述[J]. 浙江电力, 2025, 44(11): 59-71.
Si Yaxin, Liu Xunqiang, Su Xiangjing, et al.A review of offshore wind power forecasting studies[J]. Zhejiang Electric Power, 2025, 44(11): 59-71.
[11] Liu H P, Li P S, Wang Q.Short-term prediction of wind power based on BKA-CNN-BiGRU-attention[J]. Electric Power Systems Research, 2026, 253: 112543.
[12] Zhao H B, Zhu F, Ni B L, et al.Practical continual forgetting for pre-trained vision models[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026, 48(5): 5928-5944.
[13] Khan A W, Duan J D, Nawaz F, et al.Innovative hybrid NARX-RNN model for predicting wind speed to harness wind power in Pakistan[J]. Energy Reports, 2024, 12: 2373-2387.
[14] 郑文杰,谭慧娟,赵瑞锋,等.基于ARIMA-LSTM-RBF组合模型的风机出力短期预测[J].电力科学与技术学报,2024,39(4):153-159,186.
Zheng Wenjie,Tan Huijuan,Zhao Ruifeng,et al.Short-term output prediction of wind turbine based on ARIMA-LSTM-RBF combined model[J].Journal of Electric Power Science and Technology,2024,39(4):153-159,186.
[15] Aslan A, Tasci M, Kosunalp S.Comprehensive assessment of wind energy potential with a hybrid GRU-Weibull prediction model[J]. Electronics, 2025, 14(20): 4000.
[16] Wang D, Luo Q, Gao J X, et al.Physics-constrained transformer for wind power forecasting[J]. Scientific Reports, 2026, 16: 4237.
[17] 蔡昌春, 范靖浩, 李源佳, 等. 基于TPA-MBLSTM模型的超短期风电功率预测[J]. 电力科学与技术学报, 2024, 39(1): 47-56.
Cai Changchun, Fan Jinghao, Li Yuanjia, et al.Ultra-short-term wind power prediction based on TPA-MBLSTM model[J]. Journal of Electric Power Science and Technology, 2024, 39(1): 47-56.
[18] 苏华英, 林晨, 张俨, 等. 基于改进AP聚类和双重注意力机制的区域级新能源超短期出力预测方法[J]. 广东电力, 2025, 38(3): 8-17.
Su Huaying, Lin Chen, Zhang Yan, et al.Regional-level new energy ultra-short-term output prediction method for new energy sources based on improved AP clustering and dual-attention mechanism[J]. Guangdong Electric Power, 2025, 38(3): 8-17.
[19] Zhang Z Y, Zhang J K, Zhao H L, et al.Short-term wind power prediction based on GWO-VMD-FE and TCN-BiGRU[J]. Electric Power Systems Research, 2026, 253: 112568.
[20] Wang J Z, Kou M G, Li R Z, et al.Ultra-short-term wind power forecasting jointly driven by anomaly detection, clustering and graph convolutional recurrent neural networks[J]. Advanced Engineering Informatics, 2025, 65: 103137.
[21] Pei F Q, Zhang B W, Long S Y, et al.BiLSTM-PINN-based wind turbine power prediction architecture and anomaly data identification[J]. Renewable Energy, 2026, 261: 125182.
[22] Wang R F, Wu J T, Cheng X, et al.Adaptive expert fusion model for online wind power prediction[J]. Neural Networks, 2025, 184: 107022.
[23] Wu Z Y, Fang G H, Ye J, et al.A reinforcement learning-based ensemble forecasting framework for renewable energy forecasting[J]. Renewable Energy, 2025, 244: 122692.
[24] Zhao J, Guo Y Y, Lin Y H, et al.A novel dynamic ensemble of numerical weather prediction for multi-step wind speed forecasting with deep reinforcement learning and error sequence modeling[J]. Energy, 2024, 302: 131787.
[25] Romero-Barrera A J, Sipols A E, Paricio-Garcia A, et al. Short-term wind power forecasting integrating wake effect modeling with variational mode decomposition enhanced deep learning architectures[J]. Energy Conversion and Management, 2026, 348: 120738.
[26] Xu P Z, Zhang S J, Luo P X, et al.Data-feature hybrid optimization for ultra-short-term wind power forecasting[J]. International Journal of Electrical Power & Energy Systems, 2026, 176: 111742.
[27] Wang Z, Zhang K F, Ai X, et al.A dual-optimized hybrid deep learning framework with RIME-VMD and TCN-BiGRU-SA for short-term wind power prediction[J]. Applied Sciences, 2026, 16(3): 1531.
[28] Shoaei M, Noorollahi Y.A novel approach for short-term prediction of solar and wind power production using recurrent neural networks and empirical mode decomposition[J]. International Journal of Electrical Power & Energy Systems, 2025, 172: 111384.
[29] Ouyang K C, Wei D D, Sha X Y, et al.Beaver behavior optimizer: a novel metaheuristic algorithm for solar PV parameter identification and engineering problems[J]. Journal of Advanced Research, 2026, 84: 525-555.
[30] Wu T Y, Li H N, Chu S C.CPPE: an improved phasmatodea population evolution algorithm with chaotic maps[J]. Mathematics, 2023, 11(9): 1977.
[31] Feng F, Li K C, Cai M J, et al.Hybrid hyper-heuristic algorithms for global optimization in engineering designs[J]. International Journal of Machine Learning and Cybernetics, 2026, 17(3): 130.
[32] 陈文颖, 吴梦达, 刘帅, 等. 频-时域特征解析协同多尺度滑动窗口的短期风光功率预测[J]. 动力工程学报, 2026, 46(6): 60-69.
Chen Wenying, Wu Mengda, Liu Shuai, et al.Short-term wind and solar power forecasting via integrated frequency-time feature analysis and multi-scale sliding windows[J]. Journal of Chinese Society of Power Engineering, 2026, 46(6): 60-69.
[33] 冉华军, 杨杰飞, 董家豪, 等. 基于组合预测模型的风电短期出力场景生成方法[J]. 现代电子技术, 2026, 49(8): 114-120, 127.
Ran Huajun, Yang Jiefei, Dong Jiahao, et al.Method of wind power short-term output scenario generation based on combined prediction model[J]. Modern Electronics Technique, 2026, 49(8): 114-120, 127.
[34] Huang L S, bin Jaharadak A A, Ahmad N I, et al. A CPO-optimized BiTCN-BiGRU-attention network for short-term wind power forecasting[J]. Energies, 2026, 19(4): 1034.
[35] Kheshti M, Yang H Y, Liang Y C, et al.Heterogeneous multi-agent proximal policy optimization based control of solid oxide electrolysis and wind turbine generator for fast frequency response[J]. Applied Energy, 2026, 409: 127505.
[36] Batool A, Kim Y W, Byun Y C.MAVL-DRL: Multimodal foundation using multi-agent deep reinforcement learning for intelligent predictive maintenance of wind turbine energy systems[J]. International Journal of Electrical Power & Energy Systems, 2026, 177: 111781.
[37] Li C Y, Han F, Song Y H, et al.Minute-level wind power forecasting for microgrids using hybrid deep learning and dynamic Kalman filtering[J]. International Journal of Electrical Power & Energy Systems, 2026, 176: 111727.
[38] 田书欣, 韩雪. 基于正交小波变换的LSTM-ARIMA海上风速组合预测模型[J]. 智慧电力, 2023, 51(7): 39-43, 50.
Tian Shuxin, Han Xue.LSTM-ARIMA offshore wind speed combined prediction model based on orthogonal wavelet transform[J]. Smart Power, 2023, 51(7): 39-43, 50.

Funding

National Natural Science Foundation of China (No. 51741907) and the Open Fund of Hubei Provincial Key Laboratory for Operation and Control of Cascaded Hydropower Station (No. 2025KJX05).