Seismic Wave Feature Extraction And Classification Based On Artificial Intelligence

Natural seismic wave data comes from various seismic monitoring stations,It is caused by the movement of rocks in the earth’s crust.The data of ground motion is hard-won and very precious.At the same time,ground motion plays an important role in structural design,seismic analysis and seismic study of various projects as an external excitation,Therefore,it is necessary for us to study the natural seismic waves.As an external excitation,the three elements of ground motion characteristics are amplitude,spectrum and duration.In different seismic environments,the spectral characteristics of ground motion data are significantly different.In architectural design and engineering practice,ground motion data is grouped according to site conditions as a basic consideration.However,even under the same site category,the spectrum characteristics of ground motion data still vary greatly.How to deeply understand the ground motion according to its characteristics is a hot topic.In this paper,based on wavelet transform of seismic waves and artificial intelligence technology,two new seismic wave classification methods are proposed.In this paper,seismic wave data from PEER website in the United States are collected and sorted,and then seismic wave wavelet transform is carried out.In combination with artificial intelligence technology,seismic wave features are extracted and classified.The new type of seismic wave is predicted,the model is built by using ANSYS software,and we find that the classification prediction in this paper is proved to be effective by comparing and analyzing the seismic response of the model,.First of all,we collect seismic wave data,and then conduct feature screening and Fourier transform and wavelet transform,and then compare the latter two,understand the principle and advantages of wavelet transform,and select the appropriate wavelet transform mode for seismic waves.Finally,the continuous wavelet transform is determined.Because continuous wavelet transform needs to confirm the wavelet base and the scale of transformation,in this paper,we use MATLAB software programming and analysis and compare a large number of data,then we think the mother wavelet – Morl wavelet is suitable for wavelet transform,we obtain the wavelet coefficient graph of seismic wave signal.Secondly,a certain number of representative wavelet coefficients are selected which are based on the main characteristics of seismic waves,.Feature extraction is carried out for each image including grayscale symbiosis matrix and HOG feature extraction.Finally,combining support vector machine(SVM),the intelligent identification and classification of seismic waves are carried out.By adjusting and training the algorithm,the final recognition accuracy is 95%.Then,these wavelet coefficients are formed into a picture training library,and images are recognized and classified by convolutional neural network Alex Net,and the classification accuracy is compared with that of SVM.Finally,the trained SVM model and Alex Net model are used to classify and predict the new seismic waves.The two common frame structure models in the city are selected,and the structures are modeled by the software ANSYS.Firstly,the static analysis and modal analysis of the structure are carried out,and then the new seismic waves and two random seismic waves selected from the database are used to conduct seismic action on the structure model respectively,and the seismic response data of the three seismic waves to the structure are compared,so we can verify the correctness and rationality of the classification prediction.