BP Neural Network Optimized By Genetic Algorithm And Its Application In Inversion Of Moho Interface In The South China Sea

Determining the fluctuations of the underground density interface according to gravity anomalies is an important task when using gravity data to study regional geological formations or to carry out mineral resource exploration.At present,the method of inverting the density interface by gravity anomalies mainly uses the data processing method of spatial domain and frequency domain.The density interface inversion method of spatial domain is mainly based on the Bott inversion theory,including direct iteration method,compression mass surface method and nonlinear inversion method,and the density interface inversion method of frequency domain is mainly based on Parker-Oldenburg inversion theory.After decades of development and perfection,both methods have achieved results in the inversion of gravity density interface,but in the process of inversion,the calculation of the two methods is large and the process is complex,it is difficult to add known a priori constraint information,and then the result of inversion is easy to lose some important information.In this paper,the more mature BP neural network in the deep learning algorithm is used to invert the gravity density interface,with a view to training the network by entering known information,and then achieving the purpose of a priori information constraint.This method is a feed-forward neural network with error reverse propagation,which establishes a function mapping relationship between the input and output by learning and training the known information,avoids complex calculations,and enables us to obtain the desired results more quickly.In order to quickly construct the structure of BP neural network,this paper uses genetic algorithm to screen the initial weight coefficient and threshold of BP neural network,so as to shorten the time of iteration calculation of neural network,and thus achieve the optimization and efficiency of geophysical inversion.In order to verify the effect of this method in the gravity density interface inversion,this paper verifies the method through the gravity anomaly inversion material density interface of twodimensional and three-dimensional undulating terrain model,obtains the optimized BP neural network model through the network training of theoretical positive and known interface depth information constraints,and then uses the gravity density interface inversion calculation at unknown depth,and obtains reliable results consistent with the theoretical model.On this basis,the geological-geophysical data of the South China Sea and the surrounding sea areas were collected,the Moho surface anomalies were extracted,and the network training was carried out based on the known seismic profile and the depth points revealed by the OBS measurements,which were used for inversion of the depth of Moho surface in the South China Sea and adjacent areas,and the results of the Moho surface depth of the whole area were obtained.Through the preliminary explanation of the depth change of Moho surface,a new understanding of the deep geological structure of the South China Sea and adjacent areas is obtained.