基于最小二乘支持向量机的大坝变形预测研究

Research on the Dam Deformation Prediction Based on the LSSVM Model

大坝变形预测是风险评估的关键,而涉及因素存在高度非线性.为达到好的预测效果,提出了一种基于最小二乘支持向量机(LSSVM)的大坝变形预测方法.在数据预处理方面,针对传统的参数平方、立方这种处理方式,提出变阶次概念;针对LSSVM交叉验证耗时过多,提出了一种简单可行的变参数方法 .为了快速获得优化结果,引入基于十进制的遗传算法.此外,为进一步提高预测精度,引入遗忘因子概念.最后,给出一个实例.

Dam deformation prediction is the key to the risk assessment,and the involved factors are highly nonlinear.In the paper,to achieve better prediction effects,a method of dam deformation prediction based on least squaressupport vector machine（LSSVM）is proposed. First,the traditional transaction way of the square and the cubic of thewater level is changed by introducing the concept of variable orders. Second,to solve the time-consuming problem,asimple and feasible method of variable parameters is suggested to escape the cross validation of LSSVM. Third,inorder to get the optimization results quickly,the genetic algorithm based on decimal system is adopted. In addition,the concept of forgetting factor is used to improve the prediction accuracy further. Finally,an example is given.