Research On Fault Diagnosis Of Photovoltaic Power Plant Based On Machine Learning Algorithm

Since global energy demand as well as the need of environmental protection continues to increase,photovoltaic(PV)power generation which is one of the major renewable energy technologies has developed rapidly in recent years.However,during its long-term outdoor operation,diverse environmental and climatic conditions are likely to induce operational fault of PV power station.Therefore,it is important to develop a real-time diagnostic system to detect such faults thus ensuring the safety of PV power station.Among various diagnostic methods,artificial intelligence is considered to have advantages in both technical and application prospects.Hence,several kinds of machine learning algorithm is used to solve the fault diagnosis problem of PV power station,based on the experimental on both an actual PV power station and an acceleration simulation system.The work completed in this paper is introduced as follows.In the first part,this paper investigates the output characteristics of PV power station under shading.The experimental results show that the output current of PV power station has “Markov” feature when it is shaded,which is consist with the Markov chain conditions.As a result,a Markov model is able to be developed to calculate the shading probability of the PV power station.Furthermore,the volatility analysis of the shading probability makes it possible to decide whether there is shading and what type the shading is.An one month verification experiment shows that the calculation results of above algorithm agrees well to the outdoor experimental data,indicating the feasibility and accuracy of the algorithm.In order to obtain training data for machine learning models,this paper then designs an accelerating system to simulate PV power station,since the collecting of characteristic parameters of outdoor PV system with faults is very difficult and time consuming.The feasibility of the accelerating system is verified in several aspects,which shows that it can be used to extract operational data from PV power stations in different status: normal operation,fixed external shading,cloud shading and aging of PV modules.Finally,a fault classification method for PV power stations is proposed based on support vector machine algorithm.Unlike other methods,this method tries to use time series data as the model input,because of which the distinguish of clouding shading,fixed external shading and aging of PV modules becomes possible,although they exhibit similar behavior in the decay of electrical characteristics.The training and verification of the algorithm is completed using the acceleration system and the results illustrate that our method is good in performance and accuracy.