Estimation Of Earthquake Source Parameters Based On Deep Neural Network

In recent years,the research of ground motion field simulation and fault rupture mechanism based on the estimation of source parameters has been widely concerned.Based on the current investigation,earthquake source parameters usually are estimated on the basis of the multiple ground motion recordings,while it bring computational burden.For the purpose of high calculation efficiency,some researchers estimate some source parameters such as magnitude and epicentral distance based on single station ground motion records,but the estimation error is large.In order to improve the calculation efficiency and estimation accuracy,this paper proposes a fast source parameter estimation method based on the deep learning.The main research contents are as follows:(1)Firstly,in purpose of detecting the 10.24 s time window where the seismic phase arrives,the convolution neural network of time window detection is constructed.After testing the network,the detection accuracy reaches 98.7%.and the detection accuracy reaches 98.7%.Then,in order to deeply extract the phase pick-related features this paper constructs the seismic phase pick network based on convolution neural network and bidirectional gated recurrent unit(GRU).Taking the 10.24 s time window ground motion records as the network input,the P-wave and S-wave phase can be picked up quickly and accurately.Compared with the existing phase pick methods,this method has higher phase pick accuracy,and can provide more accurate phase arrival information for the rapid estimation of source parameters.(2)The magnitude estimation network and the epicenter distance estimation network are constructed based on the convolutional neural network.In order to learn the related features of magnitude and epicentral distance estimation in time and frequency domain,three component ground motion records of a single station after wavelet packet transform are used as network input.Due to the full consideration of the influence of the seismic frequency information on the magnitude and epicenter distance estimation,the networks realize the magnitude and epicenter distance estimation with high accuracy and high calculation efficiency.(3)The convolution neural network of focal mechanism solution estimation is constructed to estimate the focal mechanism solution in Ridgecrest,Southern California.And the network is trained by the ground motion records of different focal mechanism solutions generated by FK method.In order to reasonably consider the influence of seismic record attenuation on focal mechanism solution estimation,epicentral distance is used as the auxiliary input of the network.The network only uses the ground motion records of 10 stations to realize the fast estimation of the source mechanism solution.Compared with the existing method using the ground motion records of 16 stations,the estimation accuracy is basically the same.