基于多站支路功率联合学习的分布式光伏支路异常检测方法

Abnomaly detection of distributed photovoltaic array based on joint learning of multi-station array power

近年来,分布式光伏站点数量迅速增长,频发的支路异常带来了巨大的发电效能损失,也产生了如何精准且高效检测多站支路异常的需求。为解决上述问题,提出了基于多站支路功率联合学习的分布式光伏支路异常检测方法。该方法通过多个光伏站支路异常检测任务联合学习的方式,提取了辨识多站支路异常特征的相似性和差异性表示;通过构建的多尺度卷积神经网络有效捕捉了多站支路功率中存在的差异性异常特征;利用辅助任务充分学习多站支路异常辨识特征的相似表示;采用多阶段训练策略减少辅助任务对多站支路异常检测精度的消极影响。最后,通过收集的多站支路功率数据进行实验对比,结果证明了提出方法能有效提升分布式光伏各站支路异常检测的精度。此外,该方法仅需构建一个模型即可检测分布式光伏多站支路异常,具有较好的建模便捷性。

In recent years,the number of distributed Photovoltaics(PV)stations has increased rapidly.Frequent PV array anomalies have caused a great loss of power generation efficiency,which brings the demand for detecting multi-station PV array anomalies accurately and efficiently.To solve this problem,an anomaly detection method based on the joint learning of multi-station PV array power was proposed.In this method,the similarity and difference representations of PV array anomaly identification features were firstly extracted with the joint learning of array anomaly detection tasks of multiple PV stations.The multi-scale convolution neural network was constructed to capture the differential anomaly identification features of multi-station PV array power.Then,the auxiliary task was used to fully learn the similar representation of PV array anomaly identification features.A multi-stage training strategy was adopted to reduce the negative impact of auxiliary tasks on the accuracy of PV array anomaly detection.In the comparison of multiple experiments,the proposed method had a great performance in improving the accuracy of array anomaly detection on multiple distributed PV stations.In addition,this method also had the superiority in modeling convenience,because only one model needed to be built to realize the anomaly detection of distributed PV multi-station arrays.