Study On The Methods Of Multivariate Analysis And Computational Intelligence In The Optimization Of Spatial Variables

In the field of geoscience, geological features in spatial, physical and chemical aspect related to geological phenomena and processes are usually represented by geoscience variables.Geoscience variables have three characteristics: time, spatial and attribute.Any geological model is supported by geoscience variables and their relationship.As the main branch of digital geoscience applications, prediction of mineral resource has become increasingly concerning.As a result,many mathematical methods and techniques in geology have appeared.The research on geoscience variable and model is focused on the construction of linear or non-linear mathematical model and the measurement of variable relationship.The selection and assignment of geoscience variable is a critical task in the work of mathematical model construction.The raw material of geoscience variable comes from geological, geophysical, geochemical and remote sensing data.Under normal circumstances, the raw material is not the same as geoscience variable.In order to generate the geological variable parameters of the model,the raw material must be processed,transformed and converted.Therefore,the optimal selection and conversion of geological variable is a key procedure to improve the reliability and effectiveness of mineral resources prediction model.In conclusion, for mineral resources prediction,this paper discussed the structure,association, and optimal selection of geological variable in complex and multiple geoscience data.The main research was to optimize and compress complicated geological variable, with emphasis on innovating feature extraction and feature selection methods associated with the category of dimensionality reduction, in addition to combining the multivariate statistical model, artificial intelligence and intelligent evolution model, and integrating the advantages of each method.Ultimately,the simple and precise set of geoscience variable was formed,which provides an important basis for research on how to improve the precision and accuracy of mineral resources prediction.The studies included:(1) The observation data of geochemical element and geochemical anomaly in study area was selected as research data.The combination of elements that is directly related to mineral resource was extracted as a direct mineral anomaly identification sign. Simultaneously,the target geological variable parameters in predicting model were determined.(2) In interdisciplinary original data, most of which came from geology,geochemistry, geophysics and remote sensing, the principle of optimal selection was proposed.Finally,the selected geological variable contributed to the prediction model of minimal resources.In this paper, the project named "Goldmine, multi-element metal mineral resources prediction in 1/200000 scale in Ji Lin Bai Shan" was taken as an example.The research and innovations include:(1) PCA-FA-CA-BP algorithm1) Geochemical variables were spatially transformed through PCA-BP hybrid algorithm based on principal component analysis and Factor Analysis Model.As a result,complex variables by a linear transformation were expressed as a low-dimensional integrated factor.Mutual correlation between variables in the combined variable measured by linear correlation analysis theory was helpful to explain geochemical anomalies reasonably.2)Using geochemical anomaly as the main input factor, along with rock geochemical element,geographical landscape,soil environmental parameters and other factors are totally taken as neuron of input layer,the BP Neural network integrated anomaly identification model was constructed.The geochemical anomaly extraction and geochemical variables optimization process were expressed as a nonlinear dynamical system.By learning from the sample data, an intelligent classification identifier model can train and produce itself.This intelligent model breaks certain restrictions of traditional mathematical statistical model. The geochemical anomaly recognition based on BP neural network as a "black box" identification model had ahigh classification accuracy.(2) IQ-QT algorithmFor geoscience variable optimization selection problem in the mineral resources prediction model, this paper proposed two hybrid algorithms.One of the algorithms is IQ-QT algorithm. First, this algorithm calculated the amount of information that expressed each candidate variable's contribution to mineral resource.Next,the initial selection of geoscience variables was done based on the amount of information.Finally, based on the idea of successive regression and the "Reduce Project Method" which came from the quantification Theory I model,contribution of variables was calculated, variables were repeatedly introduced and removed,and the best variables were selected ultimately.(3) IQ-GA-BP algorithmThe other one of algorithms is IQ-GA-BP algorithm.The only difference with IQ-QT algorithm is that the genetic algorithm was used to select optimized variables in the secondary extraction.In other words,the geoscience variable was mapped to biological individual by encoding technology.Then by individual biological evolutionary selection, crossover and mutation mechanism,the best combination of variables was selected in a population from generation to generation in the evolution.