Research On The Anomaly Extraction Based On CNN And PSO-DLS Inversion Of The Down-Hole TEM Method

In China,coal mine production has gradually entered the deep mining stage,the threat of water disaster is more serious,and the problem of insufficient accuracy of ground transient electromagnetic detection is also emerging.In this paper,the research on the down-hole transient electromagnetic detection technology is carried out in order to improve the detection accuracy of coal mine water hazards.By constructing a typical three-dimensional geoelectric model,the three dimensional forward modeling method is used to simulate the three component response characteristics of typical water bearing abnormal bodies.Based on the convolution neural network model(CNN),the signal extraction of the abnormal field in the down-hole transient electromagnetic detection is realized,and the particle swarm optimization damping least squares(PSO-DLS)combined algorithm research,data inversion processing and engineering example analysis are carried out.The main achievements are as follows:(1)The three dimensional finite difference time domain algorithm is used to simulate the transient electromagnetic response of a typical geoelectric model,and the characteristics of the transient electromagnetic response of different water bearing abnormal bodies are summarized;(2)Based on the statistical analysis of apparent resistivity curves of nearly 800 boreholes,the general electrical characteristics of coal bearing strata in Carboniferous Permian and Jurassic are summarized;(3)In the data processing,the deep learning theory is introduced,the convolution neural network algorithm is adopted,the CNN multi-layer architecture model is constructed,and the progressive learning strategy is designed.Through the step-by-step learning method,the training learning from the total field signal of the down-hole transient electromagnetic response to the abnormal field signal is realized,which solves the problem that it is difficult to extract the abnormal field signal of the down-hole transient electromagnetic response;(4)In order to solve the problem that the conventional linear algorithm needs initial value and the non-linear method is inefficient in data inversion,a PSO-DLS combined inversion algorithm for the down-hole transient electromagnetic field signal is proposed,and the corresponding data processing and inversion software is developed;(5)After the theoretical model test and engineering exploration verification,the accuracy of the anomaly extraction algorithm and PSO-DLS inversion algorithm based on CNN progressive learning strategy proposed in this paper is further proved,and the validity of the inversion software developed is proved.The main innovations of this paper are as follows:(1)In order to solve the problem that the signal of the down-hole transient electromagnetic abnormal field is weak and difficult to extract effectively,a method of extracting the signal of the abnormal field based on CNN algorithm is proposed,and a convolution neural network model suitable for extracting the signal of the abnormal field is established;(2)Aiming at the problem that the training effect from the total field signal to the abnormal field signal is not good,the deep learning convolution neural network incremental learning strategy is proposed to improve the extraction accuracy of the abnormal field signal;(3)Based on the advantages of PSO non-linear inversion algorithm without specifying the initial model and DLS algorithm with high efficiency of joint optimization,PSO algorithm is improved and studied to realize the PSO-DLS combination algorithm inversion of abnormal field.