Research On Transformer Fault Diagnosis Based On LIF Technology And Optimized BP Neural Network

Transformer plays a vital role in the operation of the power system,and its working state directly affects the safety and stability of the power network.Therefore,the diagnosis and research of transformer fault types can scientifically and reasonably evaluate the working state of the transformer and discover the hidden dangers in time.Traditional methods such as dissolved gas analysis in oil and three-ratio method are not suitable for the construction of online detection system,and the detection parameters are less and the accuracy is low.In view of this problem,this paper proposes the application of laser induced fluorescence(LIF)combined with neural network in the fault diagnosis of transformers,and makes an accurate assessment of transformer fault types.Firstly,this paper introduces the types of transformer faults and common detection methods.Secondly,it introduces the application basis and working principle of LIF technology,and then constructs an experimental device for laser-induced fluorescence and obtains the spectral data of experimental samples.In this experiment,MSC and SNV preprocessing methods were used to weaken the baseline drift of the original fluorescence data sample and improve the signal-to-noise ratio of the data.PCA principal component analysis is used to reduce the amount of calculation and extract the data characteristics of the sample.BP,PSO-BP,IPSO-BP and SSA-BP classification models are established to obtain the classification results of transformer fault types.The best model for transformer fault oil classification is selected by evaluating parameters.The output results and evaluation parameters of each model are summarized as follows:(1)In the BP neural network classification model,the fault characteristics of transformer oil are used as the input of the network,and the type number of fault oil is used as the output of the network.Therefore,the topological structure is 3-13-1.The MSC-BP classification model has the best classification effect.The specific evaluation index parameters are as follows.R2=0.993,the mean relative error was MRE=0.0424,and the root mean square error was RMSE=0.0098.(2)In the PSO-BP model,the weights and thresholds are optimized according to the speed and position update of individual extremum and group extremum.The best classification result is MSC-PSO-BP model,the evaluation index parameters are as follows.R2=1,MRE=0.0015,RMSE=3*10-5.(3)In order to avoid the problem of PSO-BP model falling into local optimum,the improved particle swarm optimization algorithm is used to add mutation factors to the original model and increase the search range of particles.The classification effect of MSCIPSOBP was the best,and its evaluation index parameters are as follows.R2=1,MRE=5*10-5,RMSE=1*10-9.(4)In the SSA-BP model,the particles are regarded as the adder and discoverer in the sparrow,and the optimal value is determined according to the position update of the two.MSC-SSA-BP performed best in the model,and its evaluation index parameters are as follows.R2=1,MRE=7*10-7,RMSE=1*10-11.Therefore,in the classification model of transformer fault oil,MSC-SSA-BP has the highest accuracy,and its classification error is infinitely close to 0.The model has good convergence performance and can reduce data calculation time to the greatest extent.Compared with other classification models,it has unique advantages in solving the optimal problem and can realize accurate classification of transformer fault types.Figure[41]Table[6]Reference[63]...