Research On Power Cable Fault Online Diagnosis Based On Deep Learning

With the increasing standardization requirements of urban construction,power cables have been widely used in urban distribution networks due to their advantages such as small footprint,strong concealment,and stable power supply.However,early cable faults have a short duration and are difficult to detect in a timely manner,and their repeated occurrence will damage the cable insulation,leading to permanent faults.Therefore,it is necessary to conduct real-time monitoring of power cables to ensure their safe and reliable operation.However,during the monitoring process,a large amount of fault data has not been effectively identified,and the problem of false detection and missed detection still widely exists.Based on this,this article proposes a deep learning based online diagnosis method for power cable faults.By fully mining effective features from historical monitoring data,accurate online diagnosis of early cable faults is achieved,providing reference for establishing an effective and reliable intelligent online warning platform for cables.Firstly,by comparing the changes in cable parameters before and after the fault,the causes and mechanisms of early faults are analyzed.Based on the unique electrical signal morphology of early cable faults,a Mayr arc model is built that can fully simulate the characteristics of early faults.Through simulation,the arc characteristics of different parameters are analyzed.At the same time,two line models of 10kV cable single line and branch line are built according to the urban distribution network structure.Considering the confusion of early fault and over-current disturbance,and the diversity of early fault influencing factors,different locations and operating parameters are selected to join the fault module and over-current disturbance module to obtain a variety of fault feature data sets,making up for the shortcomings of poor adjustability and insufficient data volume of existing data sets.Secondly,for the cable single line failure,the cable failure recognition method based on multi-layer perception machine is proposed.Considering the impact of noise on the diagnostic results,it has constructed the de-noise evaluation index for the selection of the appropriate small wave base function to obtain the best de-noise effect.Extract the energy of different frequency bands through a wavelet decomposition as fault characteristics,and perform data pre-processing through various ways to verify that the energy characteristic matrix that obeys the normal distribution is more conducive to fault diagnosis.Subsequently,a multi-layer perceptual machine failure diagnostic model was established,and the optimization strategy was added during the network training process to avoid overfitting phenomena.T-SNE visualization technology compared the diagnostic model with traditional machine learning,which effectively solved the network.Diagnosis of poor visualization.The results show that the diagn ostic results of the model are more concentrated and stable,and it has significantly improved the accuracy of fault diagnosis compared to traditional machine learning.It has verified the effectiveness and reliability of the model in the diagnosis of single-line cable failure.Finally,for the cable branch line failure,in the basic convolutional neural network model,gradually introduce long-term memory network modules and channel attention mechanism modules to design a deep learning combination model for cable failure diagnosis.The diagnostic model not only enhances the expressiveness of the fault characteristics in the time dimension,but also enhances the contribution of key features to the diagnostic results through the channel attention mechanism.After repeated adjustment of the network structure and parameters,the model is verified that the model is at different sampling frequency frequency Accuracy of fault diagnosis.The results showed that the weight distribution of different faults was distributed through the channel attention mechanism,and the accuracy of the fault diagnosis of the combined model was higher than the basic model.In addition,the model still has good network stability,generalization and robustness in the low-frequency sampling and noise environment.