Research On Life Prediction Method Of Electromagnetic Relay Based On Improved DBN And Softmax Regression

As an important basic electrical component,electromagnetic relays exist in various electrical connections,and play an extremely important role in circuit switching.Their reliability and lifespan directly affect the safe and stable operation of the entire system.Therefore,after the relay is produced It is necessary to carry out long-term reliability life test before going out.Traditional relay reliability life experiments are performed by professionals.The experimental data produced by the experiment is very small and may have errors.Because the relay is a basic electrical component,it is in great demand in both civil and military fields.However,in practice In the production environment,the life prediction of large quantities of relays can only be tested using random sampling methods,and cannot guarantee the accuracy and reliability of the life prediction of each relay.Accurate life prediction has always been the research focus and difficulty of scholars at home and abroad.Deep learning is the most effective artificial intelligence technology that has emerged in the past two years and has been widely proven.It has achieved great success in the fields of pattern recognition and regression prediction.This paper proposes the application of electromagnetic relay life prediction method based on deep learning technology,which provides new ideas and implementation methods for improving the reliability index of electromagnetic relays and other components.The main research work is as follows:1.In order to analyze the data of each life parameter of the electromagnetic relay,use professional relay life reliability life experiment equipment to record each experimental life parameter of the electromagnetic relay,then visualize the obtained life parameter,and use the improved new information Based on the noise reduction process of the Mann filter algorithm,the Principal Component Analysis(PCA)based on unsupervised learning is used to reduce the dimensionality of the data,and finally the extreme gradient lifting algorithm(XGBoost)in the field of machine learning Score and rank the importance of life parameters of electromagnetic relays.2.Due to the nature of electromagnetic relay life data for variable time series,a multivariate time series prediction method based on deep confidence network(DBN)and Softmax regression is proposed.First,the pre-processed relay life data is divided into training sets and The test set,then input the training set data to each hidden layer in the DBN model for feature extraction,and finally input the obtained feature value to the Softmax regression layer for prediction and output the predicted value of life.3.Through the use of DBN to predict the life of the electromagnetic relay,it is found that the DBN mainly predicts the macro distribution of the data,and lacks the connection to the data context.For this problem,the BLSTM is used to improve the DBN,and the improved DBN based on the BLSTM is proposed.Relay life prediction method,and use the training set data to train the two-way long and short-term memory network to make the network model fully extract the time series features of the data and then input the extracted time series features into the deep confidence network model for training to make the model Learn the overall distribution of time series features,and then use the Softmax regression layer to make regression predictions on the data.In order to fully improve the prediction performance of the model and solve the problem of slow model convergence,this article selects the hyperparameters of the model through sufficient comparison experiments,and uses evaluation methods such as test data sets and root mean square error(RMSE)to evaluate the performance of the model,The model's test set accuracy rate reaches 95.5%,the root mean square error is reduced by 0.05,and the model prediction accuracy is improved.