Deep Learning Metallogenic Prospect Prediction Method Based On Ensemble Learning Idea

Mineral resources have always played an important role in the development of national economy.With the change of prospecting direction from shallow deposits to deep concealed deposits,the difficulty coefficient of prospecting is gradually increasing,and geologists pay more and more attention to the research of new methods and technologies of prospecting prediction.With the wide application of cutting-edge technologies related to ’ artificial intelligence ’ in various industries,researchers have begun to use deep learning algorithms to process complex and huge multi-source geological data,and achieved good prediction results.The Bawanggou gold deposit in the study area is located on the Shiquan-Shenhe detachment thrust nappe in the south of the South Qinling tectonic belt.Huanglong gold deposit,Jindoupo gold deposit,Changgou gold deposit,Liukeng gold deposit,Luming gold deposit,Baiguoshu gold deposit,Liushuping gold deposit and other gold deposits are distributed in the surrounding area,indicating that there is great prospecting potential in this area.This paper systematically collects the geological data and previous research data accumulated by Bawanggou gold mine for many years,carries out the application research of deep learning algorithm in the prospecting prediction of the gold mine,discusses the process of gold prospecting prediction method integrating various deep learning models and multi-source data,and delineates four metallogenic prospect areas.This research work has certain exploration significance for promoting intelligent prospecting methods.The main progress is as follows :(1)A multi-model multi-source data gold prospecting prediction process based on ensemble learning and deep learning algorithm is proposed.Firstly,the geological data accumulated in the study area and the previous research data were systematically collected,and the ore-forming factors in line with the metallogenic regularity of the Bawanggou gold deposit were selected.The inverse distance interpolation method and the buffer distance density method are used to convert the original data of the mine factor into data samples that are convenient for the computer to extract and identify the geological spatial characteristics,and the data samples are enhanced by the cutting method to obtain the training data sample set that meets the training of the deep learning model.Based on this sample set,a deep learning network model that meets the requirements of prospecting prediction is trained.Based on the pre-trained model,the metallogenic prediction of Bawanggou gold deposit is carried out,and five metallogenic favorable maps of the study area are obtained by referring to the integrated learning idea.Finally,four metallogenic prospect areas are delineated.(2)The idea of ensemble learning is introduced to improve the stability of prediction results.In this paper,five convolutional neural network models and visual self-attention mechanism model are trained using the training sample set,and the pre-trained prospecting prediction model is used for metallogenic prediction.The results show that although the prediction accuracy of a single network has reached the prediction requirements,the prediction results of different network structures are quite different,showing the instability of the prediction results.With reference to the idea of ensemble learning,five combination methods are used to combine the strategies of deep learning algorithms,and finally five mineralization favorableness distribution maps are obtained.The research shows that the combination of six deep learning model algorithms by ensemble learning method increases the stability of the prediction results,and also increases the reliability of the metallogenic prediction method based on deep learning algorithm.(3)Four metallogenic prospective areas are delineated.The No.1 metallogenic prospect area is located in the Shibangounao-Tiefosi-Shaojiagou area in the north and middle of the study area.The area is controlled by multiple brittle faults and developed structures.The No.II metallogenic prospect area is located in the Xiejiapo-Mao ’ergou area in the northwest of the study area.The surface of this area is exposed to the middle section of the second lithologic section of the Silurian Meiziya Formation,which is passed by a number of near northwest-south brittle faults.The No.III metallogenic prospect area is located in the Caihuagou-Tuguanzi-Tuzhaizi area in the southern part of the study area.The area is scattered in the north of the RF4 ductile shear zone and is controlled by F-3,F-4 and F-5 brittle fractures.The No.IV metallogenic prospect area is located in the south of Xiangzipa in the eastern part of the study area.The area is controlled by the F-3 brittle fault,and the stratum is the junction of the upper and middle sections of the Lower Silurian Meiziya Formation.