Identification Of Geochemical Anomalies Associated With Femineralization Based On Machine Learning Methods

Hunjiang region of Jilin Province is an important base of mineral resources,which is rich in iron mineral resources.At present,more than 20 iron ore spots have been found.In this paper,we based on regional geological survey,geological research and regional geochemical exploration data,using the machine learning in the isolated forests,one class support vector machines and random forests and cuckoo of swarm intelligence algorithm,bats,longicorn algorithm and ant colony algorithm model,building 5 kinds of comprehensive geochemical anomaly identification method,including: Max-minimum ant colony algorithm model based on small perturbation,reverse cuckoo optimized isolated forest model,adaptive differential bat optimized single-class support vector machine model,competitive longhorn beetle optimized random forest model and hierarchical clustering,singularity analysis and self-organizing feature mapping neural network model.The above five models are all applied to the geochemical data of stream sediments in the study area of 1? 200,000 to identify and extract the geochemical anomaly information related to iron mineralization in this area.The application results show that: The five kinds of integrated model in geochemical anomaly information recognition in the study area are achieved good effect,and enrich the jilin Hun Jiang regional geochemical anomaly identification method system,ultimately for ore-prospecting work in jilin Hun Jiang regions provides a more effective means of technology,iron mineral resources prospect of the further provide certain scientific basis for target prediction and the instruction significance.1.A reverse-learning cuckoo optimized isolated forest model is constructedIn the past studies of geochemical anomalies,the artificial trial-and-error method is usually used to determine the optimal parameter combination of isolated forests,but it is difficult to find the global optimal parameter by the artificial trial-and-error method.The reverse learning cuckoo isolated forest model utilizes the ability of cuckoo search algorithm in finding the optimal path and solving the optimal solution to achieve the purpose of automatic parameter optimization,and utilizes the search performance and the ability to obtain the global optimal solution to improve the performance of the overall model.2.A single-class support vector machine model for adaptive differential bat optimization is constructedThe model utilizes the bat algorithm based on differential evolution and adaptive improvement combined with one class support vector machines,in the process of using differential evolution algorithm in each iteration produces a better solution,algorithm of adaptive algorithm was used to optimize the bat velocity formula and difference algorithm is the process of mutation and crossover,achieve the goal of ascension convergence accuracy and speed,so that the entire model performance improvement.3.The competitive longicorpus longicorpus optimal random forest model is establishedThe model has the characteristics of small operation scale and high efficiency,which is combined with the random forest model to realize the automatic searching of the optimal parameter combination of the random forest.Then the competition mechanism is introduced to make the model give up the individuals with low fitness value and balance the total number of the population,so as to avoid the local optimal solution and improve the performance of the whole model.4.A multi-element geochemical anomaly recognition model combining hierarchical clustering,singularity analysis and self-organizing feature mapping neural network is establishedHierarchical cluster-singularity analysis-self-organizing feature mapping neural network model is used to analyze the clustering characteristics of ore-forming elements,and the singularity analysis method is used to enhance the information,and finally the self-organizing feature mapping neural network is used to complete the identification of geochemical anomalies.In the aspect of geochemical anomaly recognition,the comprehensive model overcomes the shortcomings of traditional methods,such as neglecting spatial data structure features and requiring manual processing of relevant features,and carries out systematic clustering and information enhancement at the data level,which ensures the accuracy and stability of anomaly recognition results of selforganizing feature mapping neural network model.5.A max-min ant colony algorithm model based on small perturbations is constructedThe minimal perturbation maximum-minimum ant colony algorithm model is mainly based on ant colony algorithm.By introducing "maximum-minimum" mechanism and "small random perturbation" mechanism,the iterative efficiency can be enhanced and premature convergence can be avoided.On this basis,the global optimal effect can be achieved quickly.In practical application,indexes such as ROC curve,AUC,Youden index and the percentage of the number of mines in the abnormal area in the total number of mines in the study area are used to measure the anomaly recognition performance of these five models.The calculation results show that: In the aspect of multi-element geochemical anomaly recognition,the hierarchical clustering-singularity analysis-self-organizing feature mapping neural network model has the best overall performance,and the adaptive differential bat optimization single-class support vector machine model is the second.In the aspect of single element geochemical anomaly recognition,the max-min ant colony algorithm model based on small disturbance has also achieved good results.All the above five models can provide techniques and methods for the identification of geochemical anomalies related to iron mineralization and resource prediction in the study area.