Inversion Of Electrical Resistivity Imaging Data Based On Artificial Neural Network And Its Application In Frozen Soil Monitoring

Electrical Resistivity Imaging is based on the difference of resistivity between rocks and ores.By observing and studying the difference of resistivity between underground media,it studies the distribution characteristics and variation rules of electric field in space,so as to find out the underground geological structure and look for the underground electrical heterogeneous body.Electrical Resistivity Imaging is essentially a branch of direct current method,is developed on the basis of conventional resistivity exploration of a geophysical prospecting method in detecting section on the layout of multiple electrode at the same time,by the artificial power conduction current into the ground,underground to form stable current field,through the automatic conversion,the cross section to deploy the transposed to automatically measure and record,won the geoelectric section.Electrical Resistivity Imaging is an effective method for stratigraphic classification and detection of hidden structures,karst voids and geological landslide.It is not only widely used in the exploration of underground resources,but also widely used in engineering.However,the inversion of Electrical Resistivity Imaging is actually a very complicated nonlinear problem.In the traditional inversion method,the nonlinear problem is always linearized.Linearization of the nonlinear problem near the initial model can easily lead to the problem of local minimum and dependence on the initial model selection,at the same time,the solution of partial derivative matrix in two-dimensional finite element inversion is often difficult.With the continuous improvement of the performance of the calculation instrument and the continuous improvement of the artificial neural network algorithm,due to its unique learning memory and nonlinear approximation ability,this paper attempts to use the BP neural network algorithm in the artificial neural network algorithm to process the electrical data.The BP neural network has the advantage of being able to avoid over-reliance on the initial model,and does not need to linearize the nonlinear problem,which can largely solve the problems encountered in conventional inversion.The nonlinear inversion of neural network can also obtain a finer structure than the traditional inversion method.Once the training is completed,its efficiency and precision are unmatched by traditional inversion methods.Under the background of global warming,the global temperature rise has a great impact on the frost heaving and melting of frozen soil.The frost heaving and melting of frozen soil have brought great influence on human life and production.China is also a large frozen country,especially in the northeastern region,which is widely affected by seasonal frozen soil and permafrost.The transportation facilities,buildings and municipal works in the northeast region are seriously affected.Dealing with the impact of poor frozen soil may cause major disasters and bring huge hidden dangers to national economic construction and people's lives and property.Therefore,for the first time,this paper uses Electrical Resistivity Imaging to monitor seasonal frozen soil for a long time,and uses neural network algorithm to process high density apparent resistivity data for the first time,carries out in-depth research,understands its freezethaw law and grasps its characteristics,which has important guiding significance for us to carry out engineering construction.In this paper,BP neural network is used for inversion of high-density resistivity data.The modeling method is to train the neural network with all apparent resistivity data and true resistivity data as the training samples of the neural network,and the training samples are obtained by forward modeling based on finite difference method.Training samples in the training model of electrode number set to 36 electrode,Wenner device for the gear form,electrode spacing is set to 1 m,for 10 layers with 195 apparent resistivity data,at the same time generate number of true resistivity data for 1960,received 95,including layer? block model,training model for training the neural network.In this paper,using the classical geological model to test the trained BP neural network,the BP neural network inversion results show that the relative traditional inversion method,using BP neural network to the high density resistivity data inversion can be more elaborate geoelectric structure,and as soon as I complete my training,inversion is quick and simple.Finally,based on the corresponding prior information,this paper specially designed the training model for the measured data collected from the air-raid shelters in the Chaoyang Campus of Jilin University,and inverted the measured data with BP neural network.The inversion results show that the training is used properly.The BP neural network can invert the measured data of apparent resistivity and obtain more accurate inversion results than the traditional method.It can better describe the fine structure of the inversion target,which is not only more morphologically accurate,but also the resistance in the result.The rate is also relatively accurate compared to traditional inversion methods.Then,the paper uses BP neural network algorithm to process the measured data of permafrost monitoring in Changchun,which lasts for seven months and spans three seasons.Then,the paper uses BP neural network algorithm for the first time to process the measured data of frozen soil monitoring in Changchun,which lasts for seven months and spans three seasons.The inversion results show that under the condition of adequate training of BP neural network,The BP neural network can better reflect the whole process of the whole soil from freezing to freezing to thawing,and better reflect the freezing and thawing and electrical profile characteristics of seasonal frozen soil.