基于概念漂移监测与增量更新机制的超短期风电功率在线预测  

Online Ultra-short-term Wind Power Forecasting Based on Concept Drift Detection and Incremental Updating Mechanism

作　　者：潘春阳 文书礼 朱淼[1] 侯川川 马建军 孔祥平 PAN Chunyang;WEN Shuli;ZHU Miao;HOU Chuanchuan;MA Jianjun;KONG Xiangping(Key Lab of Control of Power Transmission and Conversion,Ministry of Education(Shanghai Jiao Tong University),Minhang District,Shanghai 200240,China;State Grid Jiangsu Electric Power Co.,Ltd.,Research Institute,Nanjing 211103,Jiangsu Province,China)

机构地区：[1]电力传输与功率变换控制教育部重点实验室(上海交通大学),上海市闵行区200240 [2]国网江苏省电力有限公司电力科学研究院,江苏省南京市211103

出　　处：《中国电机工程学报》2025年第6期2133-2144,I0008,共13页PROCEEDINGS OF THE CHINESE SOCIETY FOR ELECTRICAL ENGINEERING

基　　金：国家电网有限公司科技项目(5500-202218122A-1-1-ZN)。

摘　　要：高精度风力发电出力预测可为风电优化运行决策提供可靠依据,可提高风电的经济效益,增强风电消纳水平。然而,目前风电功率预测模型在完成离线训练后,往往很少在现实场景中优化迭代,尽管有部分研究对自适应模型进行研究,但仍缺乏针对模型在线优化的探讨,难以满足风电功率快速精准调节需求。该文基于概念漂移监测与增量更新机制,提出一种结合风力发电波动性识别与预测模型实时优化迭代的超短期风电功率在线预测方法。首先,基于历史风电场数据,利用对冲深度学习算法搭建双通道对冲循环神经网络作为预训练模型;其次,在现实的风电功率预测场景中,通过概念漂移监测算法捕捉发电序列中数据的分布变化,分析风力发电的波动性;最后,利用基于对冲算法与在线学习的增量更新机制,对预测模型进行优化迭代,对模型中每个模块的权重进行实时调整,增强模型对于波动场景的适应性。通过真实场景仿真模拟,相较于传统的离线预测模型,该文所提方法能更好地适应现实风电快速波动场景,有效提升风力发电预测的精度与准确性。High-precision wind power output prediction can provide reliable basis for optimizing wind power operation decisions,thus enhancing economic benefits and integration level.However,current wind power forecasting models rarely undergo optimization and iteration in real-world scenarios after completing offline training.Although some studies have explored adaptive models,with limited research on discussion on model online optimization,which hinders fast and accurate regulation in wind power generation.In this paper,an ultra-short-term wind power online prediction method including wind power fluctuation recognition and real-time optimization iteration of the forecast model is proposed based on the concept drift monitoring and the incremental update mechanism.First,a dual-channel hedging recurrent neural network is built as a pre-trained model using the hedging deep learning algorithm based on historical wind farm data.Then,in the realistic wind power prediction scenario,the concept drift monitoring algorithm is implemented to capture the data distribution changes in the power generation sequence to analyze the fluctuation of wind power generation.Finally,using incremental update mechanism which is based on hedging algorithm and online learning,the prediction model is optimized online,and the weights of each module in the model are iteratively adjusted in real-time to improve the model adaptability to fluctuation scenarios.Through simulation and analysis of real scenarios,compared with traditional offline prediction models,the proposed algorithm can better adapt to the rapid fluctuation scenarios of actual wind power,and effectively improve the accuracy and precision of wind power prediction.

关 键 词：在线学习 对冲算法 概念漂移监测 超短期预测 风电功率预测 

分 类 号：TM71[电气工程—电力系统及自动化]