基于改进的偏最小二乘法的防渗帷幕防渗预测模型研究

Seepage Prevention Prediction Model of Water-tight Screen Based on Improved Partial Least Squares Method

针对偏最小二乘回归法含有全部自变量引起的预测误差问题,对偏最小二乘回归法进行了改进,采用主元选择的GA-PLSR法,即引入逐步回归方法中挑选和剔除因子的思想来选择与因变量相关性较强的自变量主元,然后利用偏最小二乘回归法进行建模,再采用遗传算法对其回归系数建立目标函数进行优化,确立最后的拟合模型用于因变量的预测,并通过实例应用,将选择主元的偏最小二乘回归模型、常规的偏最小二乘回归模型及基于主元选择的GA-PLSR模型的预测结果进行比较。结果表明,基于主元选择的GA-PLSR模型的拟合效果较好,且预测精度更高。

Aiming at the problem of prediction error caused by adding all independent variables in partial least squares regression method, the idea of selecting and eliminating factor of stepwise regression method is used to select pivotal ele- ment which have a strong correlation with the dependent variable. And then partial least squares regression is adopted to establish the model. At the same time genetic algorithm is applied to build the objective function for optimization its re- gression coefficients. Finally, the fitting model is obtained to predict the dependent variable. Comparison of improved partial least squares method, conventional partial least squares method and partial least squares method based selection of pivotal element, example results show that the improved partial least squares method has good fitting effect with higher prediction accuracy.