Diagnosis Method For Multi-source Faults In Photovoltaic Power Generation System

When fossil fuels such as coal and oil are frequently in a hurry,more and more countries are beginning to pay attention to solar energy resources.Photovoltaic power generation systems have complex intermal structures and often fail in harsh environments.The monitoring methods and fault diagnosis technologies of photovoltaic power generation systems have become particularly important.Simple monitoring and techniques relying on pure manual fault detection are not only time-consuming but also inefficient.Photovoltaic power systems are highly susceptible to external environment and have complex nonlinear problems.Therefore,it is difficult to obtain the mechanism model of power plant operation in a targeted manner.In the era of rapid development of big data and artificial intelligence,the fault diagnosis method using artificial intelligence can better conform to the trend of technology development and realize the intelligentization and informationization of photovoltaic power generation system.When the photovoltaic system is switched between different states,the input and output of each component will change.Based on the structural characteristics of photovoltaic power generation system,this paper analyzes and summarizes the shortcomings and difficulties of the fault diagnosis method in the literature,and proposes a clustering fault diagnosis method.The collected data of photovoltaic power generation system is collected by DTW+k-means method.Perform a fault type diagnosis.Most of the experiments in the literature are through simulation software and eannot represent real photovoltaie power plants.The micro-photovoltaic power plants manufactured by the team were used to conduct experiments in indoor and outdoor real environments to verify the effectiveness of the method.The main research contents and innovations of this paper are as follows:1.Firstly,the background and development history of photovoltaic power generation system are introduced,and the types and diagnostic methods of photovoltaic faults in domestic and foreign literatures are summarized.Because a large amount of tag data cannot be obtained in the operation of a real photovoltaic power station,the existing neural network-based fault diagnosis method does not have a large amount of prior knowledge to establish a reliable model,so such 11methods have certain limitations in practical applications.2.Photovoltaic power generation systems can provide massive time series data.When the system switches between different states,the input and output of each component of the system will change with time,so the correlation of data before and after is particularly important.Most of the methods in the literature ignore the time stamp of the data.This paper extracts the similarity of the data before and after using a dynamic time warping method(DTW).On this basis,a time series k-means fault diagnosis method is proposed.The extracted data features are clustered by k-means method.At the same time,there is no prior knowledge for the k-means method,and the problem of clustering effect cannot be judged.The intermal and extermal evaluation criteria of clustering are introduced.Cross-validation of internal evaluation criteria and external evaluation criteria to improve the accuracy and fault diagnosis performance of the clustering model.3.Since most of the methods in the literature are simulated by software such as SIMULINK9or simple research on a separate photovoltaic panel,the experimental equipment and the experimental environment do not comprehensively consider the operating state of the entire photovoltaic power generation system in the real environment.Therefore,the self-made photovoltaic power station simulation platform was used in the experiment,and experiments were actually taken place in both indoor controllable and outdoor environment respectively.The existing methods were compared with the proposed method.The results show that the time series k-means can better classify faults.