Study On Gravity And Magnetic Geological Interpretation Methods Based On Deep Learning

With the increasing processing ability of big data,new information processing technology methods are emerging,and Artificial Intelligence(AI)technology based on massive and complex data analysis has been developding rapidly.In the field of geological survey,the use of AI technology for geological exploration and research has become a research hotspot in recent years.In China,with decades of geological surveys and mineral resource explorations,a large amount of data and information on geology,geophysics,geochemistry and remote sensing have been accumulated,laying a solid foundation for intelligent analysis of geological data.This paper focuses on researching the geological interpretation method of gravity and magnetic data based on machine learning and deep learning in the Benxi-Xiuyan area,Liaoning Province,China.It adopts traditional unsupervised model analysis and deep learning-based intelligent analysis methods to explore and investigate the recognition of mining and fracturing structure intelligence in regional geology and deep geological structure characteristics.This research broadens traditional analysis using single gravity anomaly data or aeromagnetic anomaly data.It constructs a set of deep learning-based geological intelligent analysis methods for gravity-magnetic data,which provides new ideas for applying AI in the processing and interpreting of gravity-magnetic data.The research results and insights achieved show as follows:1.A new approach called the gravity-magnetic geological analysis method was proposed based on the deep learning-based gravity-magnetic spatial structure variety features.The method is based on the second-dimensional convolutional self-coding neural network model for gravity-magnetic spatial structure feature extraction and utilizes a Gaussian hybrid clustering model for feature clustering.This method can effectively reveal the correlation between a gravity-magnetic feature and regional geological features;deep geological features and deep geological structure types.Five ring structures have been identified by adopting the new approach,including two previously unknown ring structures in Gongchangling and Nanfen.These ring structures developed along the boundary of Longgang Block and Liaoji Rift Zone.According to the classification results of spatial structure characteristics,it reveals that the Liao-Ji rift in the study area shows two structural divisions.2.This paper presents a method based on an unsupervised pattern classification model of gravity-magnetic anomalies based on feature selection,mining geological features and discovering the geological patterns implied by different gravity-magnetic anomalies superposition types and distributions.It provides a new technique of machine learning for the gravity-magnetic interpretation method.3.According to the regional geological map,extracting the fracturing control point for fracturing label date and constructing gravity-magnetic fracturing recognize model training data set in a windowed manner.This paper designed and achieved a CNN-based recognition model of hidden fault structures and a residual network-driven intelligent recognition model of fault structures based on fault strikes.Both deep learning-based fault recognition methods can effectively identify the information of hidden faults.The identification results include the strike of each faults,which can display the extension of each faults,and the relationship between faults more apparently.These results can deepen our understandings of faults developing features and regional structural frame of faults in the research areas.4.Four important fault structures were recognized through intelligent recognition of fault structures.Hence,the study area can be divided into three zones based on different fault characteristics from northwest to southeast.The basic fault pattern of the study area was then summed up as follows.The northwestern part of the study area was mainly controlled by the branch faults of Tanlu Fault extending along the lower Liaohe River,while the southeastern part is controlled by the Dandong Yalujiang fault zone with developing of NNE trending faults.There is a transition zone with development of several NE-trending faults between these two fault structural zones.The main innovation points are as follows:1.This paper presents a method based on an unsupervised pattern classification model of gravity-magnetic anomalies based on feature selection,mining geological features and discovering the geological patterns implied by different gravity-magnetic anomalies superposition types and distributions.It provides a new unsupervised technique of machine learning for the gravity-magnetic interpretation method.2.A deep learning method for geological interpretation of gravity and magnetic anomalies based on the characteristics of spatial structure changes is proposed.Based on this method,5 annular structures developed along the Longgang block-Liaoji ancient rift boundary zone were identified in the study area,and 2 annular structures with no known markers were found in Gongchangling and Nanfen.3.Two supervised intelligent fracturing structure recognizing methods have been achieved based on gravity-magnetic recognise training data set constructed through regional geological map.New fractures can be discovered by applying the two approaches in the Benxi-Xiuyan area,which also provides a better understanding of the regional fracturing structure frame for effective results.