Research On High Voltage Electrical Equipment Monitoring System Based On Partial Discharge Signal Detection

As a new type of high voltage electrical equipment,high voltage pulse device has been widely used in industrial applications.However,the generation of ultra-high voltage pulse will lead to the frequency of partial discharge of high voltage pulse equipment,affect the efficiency of the equipment,and even cause damage to the equipment.At present,the traditional high-voltage electrical equipment monitoring system based on partial discharge signal detection has the shortcomings of high cost and poor real-time performance,and cannot be directly applied to the monitoring of highvoltage pulse equipment.Therefore,this paper presents a new method of partial discharge detection based on audio signal.From the perspective of cost and real-time monitoring,the traditional partial discharge monitoring system for high-voltage electrical equipment has been studied and improved.The main research work and innovation points are as follows:(1)This paper presents a partial discharge detection based on audio signal.In order to construct the audio data,this paper collects two kinds of equipment fault sound signals during spark discharge and partial discharge during the working process of high-voltage pulse equipment.Then 56 kinds of noise signals are collected in the open environment noise data set and the mechanical noise data set.With the two kinds of equipment fault discharge sound signals and noise data set,the sound data set containing 58 kinds of sound signals is constructed,which provides data guarantee for the subsequent analysis.(2)An improved extraction method of fused characteristic spectrum is proposed in this paper.The time-frequency characteristics of partial discharge sound signal are studied with partial discharge sound data set and compared with other sound signals.The result shows that the spectrum diagram can better characterize the partial discharge sound signal.In order to preserve the more complete short-term spectrum characteristics,a Mel scale is used and the Mel spectrum is obtained through a Mel filter.Considering that the Meier filter cannot fully capture the high frequency characteristics of partial discharge sound signal,an inverse Meier scale is introduced in this paper to obtain the inverse Meier spectrum through the inverse Meier filter.Finally,the feature fusion method is used to fuse Meier spectrum and inverse Meier spectrum to get the three-channel fused characteristic spectrum,which can characterize the complete characteristics of partial discharge sound signal.(3)In this paper,a lightweight deep learning model is used to detect partial discharge signal.The traditional machine learning algorithm,traditional deep learning model,improved lightweight deep learning model and the lightweight model obtained by neural network architecture search are used to identify the partial discharge sound on the Mel spectrum,inverse Mel spectrum and fused spectrum characteristic data set constructed in this paper.Top-1 accuracy and Top-5 accuracy are used as evaluation indexes of the model.The experimental results show that all the methods have better recognition effect when identifying fused spectrum features,and the performance is the worst when identifying inverse Meier spectrum features.The improved depth learning lightweight model and the lightweight model obtained by neural network architecture search in this paper still maintain a high recognition effect even after cutting a large number of model parameters.The accuracy of Top-1 is only 0.2% lower than that of traditional depth learning model,and the highest accuracy can reach 99.53%.However,the training speed and reasoning speed of the model has been greatly improved.The improved model has better comprehensive performance.(4)In this paper,a voice separation model based on Transformer is proposed to solve the problem of mixing multiple signals in the sound data collected in industrial environment.This model introduces a new Conform model and integrates the Convolutional Neural Network(CNN)module into Transformer to improve the ability of extracting local information.At the same time,the segmented processing scheme is used to solve the problem of voice signal overlap and three subwindows are used to process the voice frame data with sliding window.Compared with the traditional baseline model,the model presented in this paper shows better performance on Libir Css data set,WSJ0 data set and the data set constructed in this paper.It also has better voice separation effect even in the case of high signal overlap rate.The global signal distortion ratio and global signal interference ratio are 14.7db and 19.2db,respectively.Compared with BLSTM,the performance of the model is improved by about 50%,compared with traditional Transformer.Performance improvement was approximately 12%.(5)A hardware design scheme of partial discharge monitoring system for high voltage electrical equipment is proposed in this paper.First,based on the analysis of actual demand,cost and safety,Zynq-7010 chip is selected as the main control chip of the main control module,and power circuit,pulse voltage drive module,partial discharge audio detection module,networking communication circuit and RS232 serial communication circuit based on the FPGA controller are designed independently.In addition,S7-1200 Siemens PLC is used for auxiliary control and data acquisition.Finally,the experimental platform of partial discharge monitoring system for high voltage electrical equipment is built and successfully applied in actual industrial production.(6)The software of partial discharge monitoring system for high voltage electrical equipment is designed in this paper.Through the software design of each hardware circuit of the GA controller in the main control module,the software of partial discharge tracking control is realized to reduce the influence of partial discharge on the working efficiency of high voltage electrical equipment.In order to meet the requirements of field control and remote control,LNMP server and deep learning server are built in local computer room,and the corresponding environment is successfully configured.In the industrial field,the human-machine interface is realized by using the industrial control screen.On the remote-control side,the background monitoring web page is designed.The two clients provide users with simple and fast operation interface,data query and other functions.Finally,according to users with different rights,provide corresponding diversified control functions.The partial discharge monitoring system for high voltage electrical equipment designed in this paper has many functions such as partial discharge sound detection,tracking control and remote monitoring.The system has many advantages such as high intelligence,strong adaptability and friendly human-machine interface,and can effectively reduce the influence of partial discharge on high-voltage pulse equipment,which can promote the wide application of new high-voltage electrical equipment in industrial field.