Research On The Identification And Interpretation Of Geophysical Exploration Anomalies Based On Machine Learning

In the process of geophysical exploration serving geological work,data interpretation is the key link.The accuracy of interpretation results directly affects the later exploration and engineering construction.However,certain geological anomalies are often difficult to be identified in the data processing results because of the weak anomalies generated.Moreover,due to the limitation of the inversion method,most of the inversion results are geophysical models with continuous change of physical properties,which correspond poorly to the boundaries of the real subsurface masses,resulting in a large subjectivity of human judgment,which eventually leads to the inaccurate identification of the boundaries between different geological units.Currently,traditional methods such as filtering,image features,wavelet analysis,wavelet transform,directional derivative,etc.are mainly used to identify and interpret geological anomalies.These methods are meaningful for guiding mineral exploration or engineering exploration in a certain region,but they are not universal and can hardly be used to directly guide mineral exploration or engineering exploration in other regions.Moreover,the identification and interpretation of geological bodies requires a high level of empirical background knowledge and relies heavily on the subjective judgment of the researcher.Machine learning is a method that studies how to use algorithms to make computers learn from past experiences,find rules from massive historical data,establish learning patterns,improve their own performance,and then identify new data characteristics or predict future development trends.In recent years,it has been widely used in the field of geophysics.Based on this,k-means clustering based on the elbow method and density peak clustering are used in this paper to analyze the inversion results of the magnetotelluric sounding of the An Shi tunnel and the inversion results of the aerial electromagnetic method of a railway tunnel in west Sichuan,to complete the categorization,identification and separation of geological anomalies;in order to improve the accuracy of aeromagnetic data analysis and interpretation of a railway tunnel in southeast Tibet,the singular value decomposition method is applied to the measured aeromagnetic data to realize the extraction of weak aeromagnetic anomalies,and to complete the analysis,identification and interpretation of geological boundaries at different depths.Finally,the interpretation results are deeply integrated with multisource heterogeneous geological information to circle undesirable geological bodies and high-risk areas,which provides an important reference and decision basis for safe construction and geological hazard prevention and control of tunnel projects.The results obtained in this study are as follows:(1)Through mutual verification and comparative analysis of the borehole data and clustering results of An Shi tunnel,it is found that the clustering is consistent with the drilling results in the classification of rock catastrophability,which indicates that it is feasible to determine the number of clusters through the elbow method and decision diagram,and also shows that it is effective to use k-means clustering based on the elbow method and density peak clustering to complete the categorization,identification and separation of geological anomalies.The application of the above two clustering methods to a railway tunnel in west Sichuan also achieved good application results,which further verified the feasibility and effectiveness of those two clustering methods.(2)After completing the clustering of electromagnetic method inversion results,the multi-source heterogeneous geological information is deeply integrated,and by verifying and comparing the borehole data with the clustering results,the catastrophability represented by different clustering results can be roughly judged and the undesirable geological bodies and high-risk areas can be circled.Engineers should reasonably allocate their attention according to the magnitude of catastrophability,and focus on undesirable geological bodies and high-risk areas.(3)In this paper,the aeromagnetic data matrix is decomposed by singular value decomposition method,and then the appropriate singular value is selected either by singular value characteristic mean method or singular value median method,or by iterative method to reconstruct the matrix to extract the weak aeromagnetic anomaly and interpret the geological structure.The study shows that the above three methods can determine the geological boundaries and their combined use can determine the extension of the geological boundaries,verifying that the singular value decomposition can effectively extract the aeromagnetic weak anomalies and realize the analysis,identification and interpretation of geological boundaries at different depths.(4)The comprehensive study and analysis show that density peak clustering can reduce the masking of useful information in the application cases of An Shi tunnel and a railway tunnel in west Sichuan,and the guidance for engineering is generally better than k-means clustering based on the elbow method.Singular value characteristic mean method,singular value median method,and iterative method,each method extracts different strengths of aeromagnetic weak anomalies,which contain different deep information and target different depths of the interpretation objects.In engineering,it is appropriate to use these three methods together to infer the extension range of each geological boundary and achieve the purpose of high precision interpretation.