Research And Application Of Improved Deep Learning Algorithm In Fault Diagnosis Of Photovoltaic Array

Due to the large consumption of traditional fossil fuels in China,the problem of environmental pollution has become increasingly serious and people have gradually realized the importance of using renewable new energy.Among all kinds of energy,solar energy is highly praised for its safety,reliability,environmental protection and pollution-free.As an important application way,photovoltaic power generation has been widely promoted and studied.Photovoltaic array has always been a core part of photovoltaic system,because the environment is mostly very harsh,long-term operation in this condition will inevitably lead to safety issues.Once part of it fails,if it can’t be handled in time,it will affect the photoelectric conversion efficiency,and even damage the whole photovoltaic power station.Therefore,it is very meaningful to find the fault and judge the type in time and take corresponding measures according to different situations.This paper consists of three parts:(1)Photovoltaic array modeling and fault data acquisition are completed.Since there is no unified public data set in the field of photovoltaic fault diagnosis,PSIM and MATLAB are selected to design the co-simulation experimental platform,which simulates five working states of normal,short circuit,open circuit,shadow and degradation,and compares with the actual measured data of photovoltaic power station,and the results verify the effectiveness of the data.(2)SSA-DBN is used for fault diagnosis of PV modules under standard conditions.Taking the voltage and current results of co-simulation as the fault data set,SSA is used to optimize the weight and bias of DBN model,and the parameters are set.The experimental results are compared with the original DBN and DCNN.It is found that the average diagnostic accuracy of SSA-DBN is 97.71%,which is 5.61% higher than the original model,and the reconstruction error is lower,which proves that SSA-DBN has high application value;(3)For PV modules in natural conditions,the improved DRN is selected for fault diagnosis.All the fault data sets are resampled by bilinear difference to obtain more fault information,and then the environment parameters are added to form the input matrix;the structure of DRN is simplified and attention mechanism is introduced to improve the ability of filtering detail information and shorten the training period;Finally,comparing the experimental results of the improved DRN,original DRN,SSA-DBN and Alex Net four models,it is confirmed that after adding environmental parameters,SSA-DBN is no longer suitable for such situations,and the accuracy rate of the improved diagnostic model reached98.86%,The probability of misjudgment is only 1.14%,and the loss in the model training process drops the fastest.The accuracy,recall and F1 score of the improved model are better than other models,which shows that the improved model has better diagnosis effect in accuracy,stability and generalization ability.