Research And Application Of Photovoltaic Array Fault Diagnosis Based On Data Driven

China's solar energy resources are very rich,which the theoretical reserves reach 1.70 billion tons of standard coal per year and the potential for its development and utilization is very broad.As an important utilization method,photovoltaic power generation has the advantages of high utilization rate,no energy storage equipment,and strong power generation capacity.China has determined that photovoltaic power generation is an important energy utilization mode,and its installed capacity and application range have been expanding.Due to the complicated production process,the harsh working environment,the short construction period of the photovoltaic power station,and the large number of equipment,the failure of the photovoltaic power station is difficult to avoid,greatly reducing the service life and seriously affect the safety and economic operation of it.Therefore,accurate and effective intelligent fault diagnosis methods are of great significance for ensuring the normal operation of photovoltaic power plants and improving the power generation efficiency of photovoltaic systems.Based on the 10 kwp rooftop photovoltaic experimental power station of North China Electric Power University,this paper explores the data-driven intelligent fault diagnosis method for photovoltaic arrays.1.According to the single diode equivalent circuit and mathematical model of photovoltaic cell,the engineering simulation model of photovoltaic cell,module and array are established.Summarizing the electrical parameters transfer rules of photovoltaic arrays under fault conditions through simulating the output characteristics of photovoltaic system under different working conditions,extracting fault feature vectors and the mapping relationship between external behavior characteristics and fault characteristics are established.2.This paper proposes an unsupervised clustering algorithm based on fuzzy C-means to classify fault samples,and the classification results are identified by the mapping relationship between the external behavior characteristics and the fault characteristics.This method effectively solves the difficulties of relying on manual prior knowledge to screen fault samples,and make the following two improvements:I)data standardization improvement.It eliminates the influence of environmental parameter changes on the output of the photovoltaic array,so that the algorithm can correctly identify the fault characteristics.II)Improvement based on Gaussian kernel function.It reduces the computational complexity,greatly improve the general applicability and cluster correctness of the algorithm.3.By introducing the PV output time series,the time series characteristics of photovoltaic output under fault conditions are analyzed,and the judgment indicators based on similarity and bias are proposed to further improve the training accuracy of the fault diagnosis model.4.The fault diagnosis model for photovoltaic arrays is trained based on the probabilistic neural network algorithm.Through actual data verification,this method can effectively and accurately identify typical faults of photovoltaic arrays.It has high engineering applicability,can realize on-line fault diagnosis of photovoltaic system,and greatly reduces system construction and maintenance cost.