Research On Signal Processing Technology Of Nuclear Explosion Infrasound Monitoring

The infrasound monitoring of nuclear explosion is one of the four important monitoring technologies for the Comprehensive Nuclear-Test-Ban Treaty(CTBT),and the signal processing technology of infrasound array monitoring is an important research content.The signal processing technology of infrasound monitoring mainly includes three parts: first,how to extract the event data from the output data of infrasound sensor;second,how to distinguish the detected infrasound signal from other infrasound signals;third,how to analyze and calculate some important parameters of nuclear explosion according to the detected data information.In this paper,the machine learning and deep learning technologies in the field of artificial intelligence were studied,and they were applied to the recognition of infrasound array monitoring signals and the calculation of nuclear explosion parameters.This paper's main work and conclusions are as follows:1?This paper analyzed the problems of the short-time window average / long-time window average algorithm in the single channel data signal detection of infrasound station,and used the algorithm to intercept the sample data from the original infrasound data.Then,a method of signal detection based on machine learning was proposed.By calculating the signal energy of different layers of wavelet packet decomposition,a2?32 dimensional feature vector was constructed,and a detection model was established by using support vector machine and artificial neural network,and the recognition performance of the two models was compared.The experimental results showed that the detection method based on SVM can detect the event signal more effectively.2?The characteristics of infrasound signal produced by nuclear explosion were studied,and the basic characteristics of infrasound signal of nuclear explosion were obtained.The characteristics of other common infrasound signals in chemical explosion,earthquake,volcanic eruption,lightning,gale and other environments were analyzed.By constructing and extracting the signal characteristics of different events,the infrasound signal characteristics database was established.3?This paper explored the application of artificial intelligence algorithms such as machine learning and deep learning to the classification and recognition of monitoring signals of nuclear explosion infrasound array.Support vector machine,standard neural network,long and short-term memory network and convolution neural network models were constructed respectively to train and recognize the traditional numerical characteristics,instantaneous frequency and spectral entropy characteristics,time-frequency image of infrasound signal,and analyze the experimental results.The learning process was improved by using sample division and generation of antagonism network to solve the sample imbalance guidance over fitting problem.4?The location and equivalent of Watusi explosion were calculated by azimuth intersection method and Los Alamos algorithm,and then the data fusion of explosion location and equivalent was carried out by combining hierarchical clustering and K-means clustering algorithm,and the error was analyzed.According to the propagation law of infrasound wave in the atmosphere,a calculation method of sound energy weight based on nonlinear traveling wave equation was proposed,and the explosion equivalent was calculated by fusion,and the two methods were compared and analyzed.In this paper,the feature extraction algorithm,machine learning model and data fusion algorithm of infrasound array monitoring signals were studied,aiming at improving the recognition ability of various infrasound signals and the accuracy of solving the nuclear explosion parameters.The proposed event detection method based on support vector machine can effectively solve the problem of single channel data signal detection;the convolution neural network model constructed had the best performance in the depth learning model,and can be applied to the identification of infrasound array monitoring signals;the proposed nuclear explosion equivalent fusion algorithm based on sound energy weight can effectively improve the accuracy of the solution.