Research On Status Evaluation And Prediction Method Of XLPE Cables In Rural Distribution Network Based On Data-Driven

Cross-linked polyethylene(XLPE)cable has become the key hub of power transmission in rural distribution network due to its advantages of good insulation performance and high mechanical strength.However,the insulation performance of XLPE cable is not always unchanged due to the influence of the laying method and the use environment.The influence of moisture and mechanical stress will lead to the damage of XLPE cable insulation and accelerate the aging rate.The aging degree is serious,which will greatly reduce the insulation performance of the cable,resulting in failure,power failure and even greater economic accidents.The regular maintenance strategy in the traditional XLPE cable insulation aging evaluation method will lead to the occurrence of over-maintenance and under-maintenance of the cable,and can not accurately and reasonably evaluate the status of the XLPE cable in time.However,the comprehensive status evaluation of the XLPE cable often depends on expert experience,which directly affects the diagnostic performance of the cable status evaluation model.Therefore,it is necessary to evaluate the status of XLPE cables,improve the accuracy of diagnosis,give accurate advice on operation and maintenance,and ensure the reliability of power supply in rural power grids.In this paper,XLPE cable is taken as the research object,from the perspective of cable condition evaluation feature selection and prediction model improvement.And based on the idea of data-driven,XLPE cable condition evaluation and prediction research is carried out.The main conclusions are as follows:(1)In order to improve the accuracy of XLPE cable status evaluation and reduce the influence of redundant features on the cable status evaluation results.In this paper,an embedded feature selection method is proposed.Based on five characteristic parameters of XLPE cable insulation aging detection,including partial discharge intensity,service life,neutral point corrosion,visual condition and cable load state,31 initial feature sets are innovatively constructed by binary coding.Then select the best feature subset based on the performance of the initial feature set under the PSO-SVM model.The result shows that the accuracy of PSO-SVM is 98%.Comparison of Whale Optimization Algorithm(WOA)and Simulated Annealing Algorithm(SA),PSO algorithm has fast convergence speed and high adaptability.The best feature subset also has better generalization ability under different classification models.The average accuracy of the test set of PSO-SVM,BPNN,DT and NB are 98%,96%,95.4% and 96.4% respectively.In the PSO-SVM algorithm,the average accuracy rate of the feature selection method in this paper compared to the PPCs and Relief-F feature selection method test set has been improved by 0.8% and 4.6%,respectively.Compared to the original feature,the classification accuracy of the feature test set selected in this paper has been improved by 1%,and the feature dimension of the optimal feature subset has been reduced by 60% compared to the original feature.Based on the analysis of actual testing cases of XLPE cables,only 4of the 24 cables with different aging degrees were incorrectly judged,with an accuracy rate of 83.3%.The feature subset selected based on the PSO-SVM feature selection method can effectively diagnose the cable status in the field environment,which verifies the effectiveness of the feature selection model proposed in this paper for status evaluation.(2)This paper constructs a XLPE cable state prediction model based on temporal convolutional network(TCN).In order to make the prediction model have better application effect,this paper proposes an improvement strategy considering that the dependency between channels cannot be quantified in TCN network expansion convolution and the class spacing between samples that are difficult to classify for XLPE cables is too small.The results show that the accuracy of the improved TCN model test set is 93%,and the loss value is 0.00331.Compared to TCN,GRU,and LTSM models,the recognition accuracy of the proposed method on the test set is improved by 3%,6.4%,and 6%,respectively.The average prediction accuracy of the training set of the improved TCN model is the highest,at 91.6%.Compared with the TCN method,the proposed method improves by 2.46%,indicating that the introduction of SE Net can enable the TCN model to obtain the importance weights of different feature channels,enhance the feature utilization ability of the model,and improve the prediction accuracy of the model.The average loss value of the improved TCN model is 0.00223,indicating that the Focal Loss function can increase the weight of cable difficult classification samples during the iteration process,making the improved model focus more on difficult classification samples,thereby improving the prediction accuracy.(3)In this paper,based on PyQt5 and Qt Designer,an XLPE cable status evaluation software system is designed and implemented.Based on the modular design idea,the functions of data import,parameter setting,operation process display and operation results of the system are realized,and the cable operation and maintenance personnel can carry out relevant data analysis on XLPE cable data,assisting the staff to complete the status evaluation task.According to the different health status of the cable,corresponding maintenance suggestions are given.