On-line Intelligent Fault Monitoring And Diagnosis Of Photovoltaic Array Based On Random Forest

In order to cope with the increasingly severe energy shortage and environmental pollution issues,the installed capacity of photovoltaic(PV)power plants has developed rapidly in recent years.As the core of energy acquisition of PV power generation system,PV arrays are often installed in outdoor harsh environment,which are prone to various faults,causing abnormal power loss,system aging,fire hazards and other safety issues.Therefore,online real-time fault monitoring and diagnosis of PV arrays is of great significance to improve the efficiency of power generation and maintenance of PV power plants.Aiming at the problems of the existing machine learning-based PV array fault diagnosis methods,such as limited generalization performance,low accuracy,difficulty in simultaneous diagnosis and location faults,an on-line intelligent fault monitoring and diagnosis system for PV arrays based on random forest ensemble learning algorithm is studied and designed in this paper.The main work of this paper is:Firstly,a string-level PV array fault detection and location method is studied.In this paper,an outlier detection algorithm for PV arrays based on the combination of "threshold method and Hampel identification method" is adopted.Firstly,the median and median absolute deviation of all string currents are calculated,then the limit area of Hampel identification method is solved,and the preliminary fault detection is carried out.Then,the discrete threshold of the PV string currents is innovatively selected to make final judgment on suspicious outliers,which improves the accuracy of fault detection and location.The experimental verification results based on simulation and measured data show that the proposed algorithm can quickly and accurately detect faults and locate fault strings.Secondly,two kinds of fault classification diagnosis algorithms of PV arrays are studied.Two integrated learning algorithms,random forest(RF)and rotating forest(Ro F),are applied to fault detection and diagnosis of PV arrays.RF only uses the PV-array voltage and each PV-string current as the fault characteristic variable of the fault diagnosis model,and uses the grid search method to optimize the parameters of the RF model via minimizing the out-of-bag error(OOB)estimation,thereby improving the fault diagnosis model.Ro F uses the Relief F feature selection algorithm to obtain fault features with large weights of importance.Then the extreme learning machine(ELM)replaces the decision tree classifier to overcome the over-fitting problem,and obtains the optimal model parameters by using the traversal method.In order to obtain enough fault data samples,the comprehensive fault simulation and experiment were conducted on the Simulink-based PV simulation system and the laboratory PV platform.The experimental verification results based on simulation and measured data show that the optimized RF and Ro F fault diagnosis model can identify the line-line faults and mismatch faults with high precision.RF is suitable for online real-time fault diagnosis of large-scale PV arrays,while Ro F is suitable for small-scale arrays with low real-time requirements.Finally,an integrated fault monitoring and diagnosis system has been developed for the existing PV array monitoring system with low sampling rate,low flexibility and low integration.The high-speed multi-function USB-1608 G acquisition card is selected as the acquisition hardware,and the acquisition card is configured and controlled by the data acquisition toolbox based on MATLAB.The monitoring interface of the integrated host computer is designed for the sampling mode setting,sampling rate setting,mode setting,channel enabling setting,etc.,and data display and processing,and achieve real-time fault location and diagnosis functions.The work of this paper has certain theoretical and practical reference value in the field of PV array fault monitoring and diagnosis.