Synthetic Information Mineral Prediction Research Of Polymetallic Deposit In Haobugao Distinct, Inner Mongolia

The mineral prediction has been more and more difficult with the aim of prospecting turned from shallow deposit to concealed deposit. And the success rate of mineral prediction has becoming more depended on the application of new theories and technologies. In order to effectively improve the prospecting results, it is important tostudy new and more effective methods for mineral resources prediction.In recent years, with the advent of the era of geological big data, its hidden enormous value of social, economic and scientific research has attracted great attention from all walks of life. The application of big data may be able to realize intelligent prospecting. So it is the direction of prospecting development in future that realizing intelligent mineral prediction and evaluation through integrating massive geological data and using methods of big data mining and knowledge discovery.Haobugao area of Inner Mongolia is located in the northeastern end of huanggangliang-ganzhuermiao metallogenic belt. There are many ore deposits in the distinct such as haobugao lead-zinc polymetallic deposit, laogenba lead-zinc deposit,aonaodaba silver-tin polymetallic deposit, xiaojingzi lead-zinc deposit point and some polymetallicmineralization points. So the distinct has superior ore-forming geological condition and the great ore prospecting potential. It has important practical significance for the exploration of mineral resources in the area.This thesis analysed and extracted metallogenic geological and geochemical anomaly information based on the multi-geological datasets of haobugao area and established a synthetic information mineral resources prediction model for skarn type polymetallic deposit. Then several machine learning algorithms were used for mineral prediction and assement. The main achievements obtained in the research work were as follows:(1)Spatial autocorrelation method was applied to the extraction of the weak anomaly of geochemical anomalies. The moran's index of global spatial autocorrelation and local spatial autocorrelation were used to identify local aggregation and anomaly patterns of geochemical data, then the geochemical anomaly was extractedfrom high value clustering and high and low values anomaly.The experimental results showed that the weak anomalies extracted from spatial autocorrelationhad a good agreement with the results of multi dimension fractal singularity index method. It provided a good idea for the application of the traditional spatial statistical methods in the geochemical data processing.(2)A semi supervised learning algorithm is used in mineral prediction for improving the efficiency of mineral resources prediction by combining with the unlabeled samples in the case of the less known ore points. A co-training semi supervised learning model which using the support vector machine and random forest as the base classifier was used for iterative training and predicting.The results showed that the semi supervised learning models can improve the prediction accuracy compared to the traditional supervised learning models.(3)A new mineral prediction model combined spatial statistic and machine learning was builded for mineral prediction considered spatial heterogeneity of geological factors. Firstly spatial clustering analysis method was used to divide the study area, through the spatial division, the subspace sample separability was improved.Then the random forest model was used to mineral prediction in each sub space. The evaluation results showed that the forecast effect of the model was better than that of the any other models.(4)Recursive feature elimination of random forest method based space division was used to evaluate the importance of ore prospecting factors in the study area. The results showed that the importance of each prospecting factor in three sub space was different, which revealed the spatial scale effect of ore prospecting factors.(6)Six favorable mineralization areas were delineated through the prediction results of machine learning model. Among them, there were two first level mineralization favorable areas, the first predicted area was menghewula-wulanhada and the second predicted area was haobugao peripheral area. And there were four second level mineralization favorable areas, the first was xinhaote, the second was gerichaganhanshan, the third was chuludabaxi west area and the last was tenigeertu.