| 引用本文: | 臧海祥,李叶阳,张越,高革命,刘亚楠,卫志农,孙国强.基于组合分解和横向联邦学习的分布式超短期风电功率预测[J].电力自动化设备,2025,45(4):45-52 |
| ZANG Haixiang,LI Yeyang,ZHANG Yue,GAO Geming,LIU Yanan,WEI Zhinong,SUN Guoqiang.Distributed ultra-short term wind power prediction based on combinatorial decomposition and horizontal federated learning[J].Electric Power Automation Equipment,2025,45(4):45-52 |
|
| 摘要: |
| 针对现有风电功率预测精度较低且未考虑多风电场数据安全的问题，提出一种基于组合分解和横向联邦学习的多风电场分布式超短期风电功率预测方法。利用自适应噪声完备集合经验模态分解获得风电功率的多模态分量，利用奇异谱分析对高频非线性分量进行二次分解，并基于近似熵复杂度量化结果对多模态分量进行重构；在横向联邦学习框架下，采用随机控制平均算法实现深度置信网络参数的更新与聚合，以获得各重构分量的预测结果；利用贝叶斯优化算法确定重构分量的叠加系数，获得最终的风电功率预测值。基于5座风电场数据进行的算例测试结果表明，该方法在考虑多风电场数据安全问题的基础上获得了更好的预测结果。 |
| 关键词: 风电功率预测 组合分解 横向联邦学习 深度置信网络 贝叶斯优化 |
| DOI：10.16081/j.epae.202412024 |
| 分类号:TM614;TP18 |
| 基金项目:国家自然科学基金资助项目(52077062) |
|
| Distributed ultra-short term wind power prediction based on combinatorial decomposition and horizontal federated learning |
|
ZANG Haixiang1, LI Yeyang1, ZHANG Yue1, GAO Geming2, LIU Yanan2, WEI Zhinong1, SUN Guoqiang1
|
|
1.School of Electrical and Power Engineering, Hohai University, Nanjing 211100, China;2.PowerChina Jiangxi Electric Power Engineering Co.,Ltd.,Nanchang 330096, China
|
| Abstract: |
| Aiming at the problems of low accuracy of current wind power prediction and not considering the security of multi-wind farm data, a multi-wind farm distributed ultra-short term wind power prediction method based on combinatorial decomposition and horizontal federated learning is proposed. The multi-mode components of wind power are obtained by using the complete ensemble empirical mode decomposition with adaptive noise, the high-frequency nonlinear components are secondary decomposed by using the singular spectrum analysis, the multi-mode components are reconstructed based on the complexity quantization results of the approximate entropy. Under the horizontal federated learning framework, the stochastic controlled averaging algorithm of the deep belief network model is utilized to update and aggregate parameters, then to obtain the prediction results of the reconstructed components. Bayesian optimization algorithm is used to determine superposition coefficients of reconstructed components to obtain final wind po-wer predicted values. Experimental results based on data from 5 wind farms demonstrate that the proposed method considering data security across multiple wind farms can obtain better prediction results. |
| Key words: wind power prediction combinatorial decomposition horizontal federated learning deep belief network Bayesian optimization |
