基于Fisher判别分析法的岩溶塌陷预测

Prediction of karst collapse based on Fisher discriminant analysis method

为了更准确地预测岩溶塌陷,基于Fisher判别分析法,首先选择内聚力、内摩擦角、覆盖层厚度、覆盖层厚度减高水位埋深、覆盖层厚度减低水位埋深5个最主要的影响因素作为判别因子;接着以桂林市20组岩溶塌陷案例为学习样本进行计算,建立相应的Fisher线性判别函数;然后利用回代估计法对上述20组样本的预测结果进行逐一检验,其正确率为100%;随后将上述预测结果与前人运用逐步判别、神经网络判别等方法得到的结果进行比较分析,认为Fisher判别分析法具有不需进行模型和参数的选择以及不受人为因素的影响等优点;最后将建立的Fisher线性判别函数用于另外的10组实际案例进行预测分析,预测结果与实际情况完全吻合。结果表明:用Fisher判别分析法进行岩溶塌陷预测,简易方便,正确率高,实用性好;该方法是岩溶塌陷预测的一种新方法。

In order to accurately predicting the karst collapse, based on Fisher discriminant analysis method, the five main factors including cohesion, internal friction angle, cover thickness, difference of cover thickness and high buried depth of groundwater and difference of cover thickness and low buried depth of groundwater were selected as discriminant factors firstly; secondly, twenty actual samples of karst collapses in Guilin City were taken as the learning samples, and the Fisher linear discriminant function was established; thirdly, the predicting results of the above twenty actual samples were tested with back substitution estimation method, and the accuracy of predication was 100% ; fourthly, the comparison of the above predication and the predication by others with stepwise and neural network discrimination methods, showed that the models and parameters had not to be selected and personal factors had no influence on the predication for Fisher discriminant analysis method; finally, the additional ten actual samples were predicted with the established Fisher linear discrirninant function, and the results were the same as the practical situation. The results showed that Fisher discriminant analysis method for predicting the karst collapse had advantages of simple, convenient, high accuracy and practicability; the method was new for predicting the karst collapse.