Anomaly Detection And Early Warning Of Photovoltaic Array Based On Data Mining

In view of the limited storage capacity of traditional fossil energy and the demand of environmental protection,clean energy such as light energy has gradually become an important support of modern energy system.In recent years,with the rapid development of photovoltaic industry in China,the installed capacity and power generation scale are becoming larger and larger.Accurate and effective anomaly detection and early warning of photovoltaic system is conducive to maintaining the safe,stable and efficient operation of equipment,and has a very positive effect on improving the economic efficiency of photovoltaic industry.This paper chooses the current measurement point data,based on data mining,carries out the research of photovoltaic array anomaly detection and early warning technology.Facing the characteristics of photovoltaic power generation process which is easily affected by illumination,temperature and other factors,this paper establishes an anomaly detection framework based on condition recognition.Based on K-means clustering,the working condition classifier is established to recognize the operation mode of photovoltaic array effectively and to calibrate all kinds of working conditions.For the current process data of photovoltaic arrays,feature sets are extracted from common statistical parameters and parameters of GMM(Gaussian mixture model)model.And then a CR-SVM anomaly classifier based on condition recognition is constructed to train and test the feature set of current data of photovoltaic arrays.Experiments verify that the normal and abnormal data can be accurately divided,and the anomaly detection of photovoltaic arrays is realized,which provides a reference for subsequent selection of normal component data to construct prediction model.Based on the data of each working condition in the stable operation state of the normal group of photovoltaic arrays,the long Short-Term Memo(Long Short-Term Memo)in depth learning is introduced to establish a time series prediction model ofthe current measurement points of photovoltaic arrays.In the process of modeling,the optimization of the model is completed by combining various parameters.Then,based on the prediction model,the alarm rules are established to realize the effective and timely anomaly warning of photovoltaic arrays.Taking the historical data of a photovoltaic power station in Hebei as an example,the validity and feasibility of the algorithm are verified.Experiments show that the anomaly detection and early warning scheme in this paper can accurately judge and warn the anomaly of photovoltaic array current data,and has great reference value for the operation and maintenance of related equipment in practical engineering applications.