Electrostatic Discharge Signal Recognition Based On Echo State Network

Electrostatic discharge(ESD)will occur on many electrical equipment,aerial vehicles,high voltage and ultra-high voltage transmission lines.Electrostatic discharge(ESD)can cause power loss and internal damage of power equipment,and also interfere with the normal operation of other equipment.How to accurately identify the ESD signal is of great significance to the stable operation of power equipment.Therefore,the identification of ESD signals will be carried out.In this paper,the electrostatic discharge signal data collected in the laboratory is used as the research object.There is a certain amount of noise in the electrostatic discharge signal,and the main form of noise is additive white noise.Firstly,the electrostatic discharge signal is preprocessed,and the initial noise reduction is carried out by berouti spectral subtraction with optimized parameters.The better noise reduction effect is obtained,and the problems of high energy target signal sensitivity and signal distortion in traditional spectral subtraction denoising are solved.The signal with good noise reduction is detected by the improved double threshold endpoint detection,which has a very good effect on the interception of the main characteristic bands in the signal,and solves the problem of inaccurate endpoint location of the ESD signal used in this paper by the traditional double threshold endpoint detection.Then,the preprocessed ESD signals are identified by the improved deep echo state network,which solves the problems of the weak mapping ability of the traditional echo state network for high-dimensional mapping and the complexity of updating the state weights of the reserve pool of the deep echo state network,and reduces the errors in the identification,and has strong recognition ability for ESD signals.Based on the above preprocessing method and recognition method,the electrostatic discharge signal collected in this paper can be accurately identified.The main work of this paper is as follows.1)A preprocessing method combining berouti spectral subtraction based on optimal parameters and improved double threshold detection is designed.The electrostatic discharge signal is denoised by berouti spectrum subtraction method.It is found that there is distortion in the signal after noise reduction.In this paper,the distortion problem is solved by optimizing the over reduction factor and the parameters of spectral line,and the noise reduction of ESD signal is completed.In this paper,the traditional double threshold method is used to determine the starting point,and then reset the threshold value of the end detection,and add a threshold T(the time range of the signal to be intercepted),so as to complete the end detection and determine the starting point and end point.The experimental results show that this method has a very good ability of filtering and endpoint detection for ESD signal,which makes the processed ESD signal more pure and has obvious characteristics.2)An electrostatic discharge signal recognition method based on improved deep echo state network is designed.In this paper,the working principle of echo state network is studied,and the deep echo state network processed by echo state network is improved.A coder is added to the deep reserve pool to transform the high-dimensional mapping into the low-dimensional mapping.After compressing the features,it is converted to the high-dimensional mapping again through a reserve pool to make its features more prominent.In this paper,the weight collection mode of deep echo state network is improved.Different from deep echo state network,the weights of all layers are collected.In this paper,only the state weights of the last layer are collected to calculate the output weights,so as to reduce the influence of useless features and reduce the recognition error.