| 引用本文: | 郑李梦千,朱利鹏,文唯嘉,李佳勇,张聪.基于多重相关性学习的风电场SCADA数据修复及其功率预测应用[J].电力自动化设备,2025,45(3): |
| ZHENG Limengqian,ZHU Lipeng,WEN Weijia,LI Jiayong,ZHANG Cong.Multiple correlation learning-based wind farm SCADA data correction and its application in wind power prediction[J].Electric Power Automation Equipment,2025,45(3): |
|
| 摘要: |
| 风电场数据采集与监视控制(SCADA)系统实测数据中的数据缺失、噪声等非理想测量工况给短期风电功率的可靠预测带来严峻挑战。为解决这个问题，提出了一种基于多重相关性学习的SCADA数据修复方案。对于SCADA实测数据中存在的数据缺失问题，提出综合挖掘多维时序数据多重相关性的数据修复方法，对缺失数据进行初步修复；设计适用于多种复杂工况的残差神经网络，对初步修复结果进行进一步精细化处理，实现精细的缺失值修复和数据去噪；以修复后的数据为输入，通过基于多头注意力机制的卷积神经-长短期记忆深度学习网络构建高可靠的短期风电功率预测模型。华中地区2座风电场实测SCADA数据的算例分析结果验证了所提方法的有效性及其在提升短期风电功率预测性能方面的应用价值。 |
| 关键词: SCADA数据修复 多重相关性 短期风电功率预测 深度学习 残差神经网络 |
| DOI：10.16081/j.epae.202412026 |
| 分类号:TM614;TP18 |
| 基金项目:国家自然科学基金资助项目(52207094,52377095) |
|
| Multiple correlation learning-based wind farm SCADA data correction and its application in wind power prediction |
|
ZHENG Limengqian1, ZHU Lipeng1, WEN Weijia2, LI Jiayong1, ZHANG Cong1
|
|
1.College of Electrical and Information Engineering, Hunan University, Changsha 410082, China;2.Information and Telecommunication Branch of State Grid Hunan Electric Power Co.,Ltd.,Changsha 410004, China
|
| Abstract: |
| The non-ideal measurement conditions such as data missing and noise in the measurement data of wind farm supervisory control and data acquisition(SCADA) system bring serious challenges to the re-liable short-term prediction of wind power. To address this problem, a SCADA data correction scheme based on multiple correlation learning is proposed. Aiming at the problem of missing data issue in the measured SCADA data, a data recovery me-thod of comprehensively mining multi-correlation for multi-dimensional time series data is proposed to preliminarily correct the missing data. A residual neural network adapted to variety complicated operating conditions is designed to further refine the preliminary recovery results, thereby realizing fine missing value correction and data denoising. With the corrected SCADA data taken as inputs, a highly reliable short-term wind power prediction model is constructed via convolutional neural network-long short-term memory deep learning network based on multi-head attention mechanism. Numerical analysis results with field SCADA data obtained from two real-world wind farms in Central China verify the effectiveness of proposed method and its application value in enhancing the performance of short-term wind power prediction. |
| Key words: SCADA data correction multiple correlation short-term wind power prediction deep learning residual neural network |
