Application Of Support Vector Machine For Aquifer Water Content Forecasting In Geophysical Prospecting

Quantitative underground water prediction provides the important basis for the rational use of underground water resources. The traditional large-scale sampling test method has the inherent limitations of high cost and low efficiency, plus the large amount of precious water was pumped and wasted. With the rapid development of geophysical technology, applying the new geophysical methods to forecast the aquifer water content has become a hot topic. Currently, the most widely used way is the statistical regression method which predicts water content by analyzing geophysical measurement data. However, current forecasting methods still have many deficiencies. For example, these methods require a large number of efforts to distinguish the key parameters from the fuzzy ones, the calculation is complex and time-consuming, and huge amount of sample data are needed. Most importantly, the prediction accuracy and generalization ability of the current statistical regression method are still questionable. Support Vector Machine (SVM) is a new machine learning technology created by Vladimir N Vapnki in the nineties of last century.By integrating the largest interval hyperplane, Mercer kernel, quadratic programming, sparse solutions and relaxation techniques, SVM has been proven to be a promising forecasting model with the strong generalization capability in various challenging application areas.This research is supported by the scientific research funding of the Integrated Geophysical Aquifer Water Content Forecasting Technology from the Chinese National Ministry of Land. First of all, on the basis of collection of domestic water content forecasting research results and support vector machines in similar instances of successful application areas, predict the effectiveness of the aquifer is used careful analysis of geophysical methods. With known water resources to a research base, through a variety of geophysical methods, and comparative test results, an application of Quaternary aquifer water content for the prediction of integrated geophysical Optimization Models are summarized. Second, obtained by correlation analysis of physical parameters, the initial selection of a strong correlation with water content as a predictive model parameter input feature vector, groundwater prediction model which use single hole discharge as predicted output value was established. After conducting in-depth analysis and comparison of the number of support vector machine technology, the technology that the prediction model for aquifer water content needed have been determined. Through the cycle of experimental methods, radial basis function was chosen as the kernel function prediction model; and through the cross-validation method, the best parameters are determined. A geophysical method based on aquifer water content prediction model is established. Finally, using Visual C++6.0 as development tools, predictive system can be achieved. Aquifer water content forecasting ability of the inspection system to promote, used Shijiazhuang Ximazhuang and Beijing Shunyi water sources as test areas, access to good predictions.The most important outcomes from our research are:â‘ by comparing various geophysical depth testing methods, the best way to identify the water layer depth is: sounding method generates better result when the aquifer is shallow; but the electromagnetic method is recommended for the deep aquifer. When carry out the field measurements for the Quaternary aquifer water content prediction, sounding method combing with the ground penetrating radar or electromagnetic plus seismic method is recommended for shallow or deep aquifer layer, respectively.â‘¡For the first time, the support vector machine is successfully applied in the aquifer water content prediction. Also, a nonlinear inverse model without pre-defined conditions was successfully established.