Research On Fault Diagnosis Of Photovoltaic Power Plant Equipment Based On ISOMPNN

With the promotion of new energy sources,photovoltaic power generation has become more widely used around the world,and the proportion of power generation has also increased year by year.Higher requirements are required for the safety and reliability of the equipment.Once a fault occurs,it will cause irreversible damage to the equipment.Therefore,timely diagnosis of PV inverter failure is a very urgent problem.Through the observation and analysis of a large number of operation and maintenance data,it can be seen that the failure of the photovoltaic inverter has a significant impact on the real-time production and operation of the power station.Therefore,the research on fault diagnosis of photovoltaic inverter based on operation and maintenance data has emerged.This paper starts with the operation and maintenance data of photovoltaic power station,explores the relationship between PV inverter equipment and actual operation and maintenance data,finds the difference between new data and historical data.According to the above two points,the incremental fault diagnosis model for photovoltaic equipment is constructed.Through the analysis of the fault data,it is found that there are dozens of dimensions related to the occurrence of the fault,and the data range of each dimension is not uniform,and the data range between the different dimension is large.In order to solve such problems,this paper adopts the method of Self-organizing Map(SOM)to reduce the dimensionality of the data,so as to select the features of the sample and SOM can maintain the topology of the original data.The original data was replaced by the mapping data,which made the data range smaller,the value became finer,the distribution was more average,and the validity was verified on the Iris dataset.However,in the generation of new faults,the traditional learning algorithm does not have the ability to classify it.It can only be added to the historical dataset to retrain the model.As the size of the data increases,the space occupied by historical data sets and model training time will also increase rapidly.Therefore,the Self-organizing Map Probabilistic Neural Network(SOMPNN)is introduced.This model combines the advantages of Probabilistic Neural Network(PNN)with simple structure,simple training and SOM with dimension reduction and extraction features.Based on the model,the incremental learning function is realized,and the Incremental Self-organizing Map Probabilistic Neural Network(ISOMPNN)is proposed.In this paper,each sample corresponds to the value in the SOM codebook vector matrix instead of the SOM prototype vector as the input of the PNN,thus ensuring that the input sample size is not too small and the data is diverse.Besides,a reasonable size SOM structure is trained for each category of data to ensure its uniformity.As for the incremental learning of new data of known categories,the method is to update the same SOM structure as the new data category,enabling it to master the new knowledge in the new data to achieve incremental learning;As for the incremental learning of new category failures,a new SOM structure should be trained for this category of failure,learning the characteristics of its data,and integrating it into the existing SOMPNN model.In the fault diagnosis of photovoltaic inverter,the effectiveness of the incremental learning of the model is verified.At the same time,it avoids the preservation of historical data sets and spends a lot of time retraining the model,which can diagnose faults in a timely and efficient manner.