基于遗传算法优化支持向量机的大坝安全性态预测模型  

Prediction model of dam safety behavior based on genetic algorithm optimized support vector machine

作　　者：谷艳昌[1,2] 吴云星 黄海兵[1,2] 庞琼[1,2] GU Yanchang;WU Yunxing;HUANG Haibing;PANG Qiong(Dam Safety and Management Department,Nanjing Hydraulic Research Institute,Nanjing 210029,China;Dam Safety Management Center of the Ministry of Water Resources,Nanjing 210029,China)

机构地区：[1]南京水利科学研究院大坝安全与管理研究所,江苏南京210029 [2]水利部大坝安全管理中心,江苏南京210029

出　　处：《河海大学学报（自然科学版）》2020年第5期419-425,共7页Journal of Hohai University(Natural Sciences)

基　　金：国家重点研发计划(2016YFC0401608);国家自然科学基金(51979175);南京水利科学研究院中央级公益性科研院所基本科研业务费专项(Y717007,Y717012,Y719001)。

摘　　要：为提高支持向量机对大坝安全性态的预测效果,提出基于遗传算法优化的GA-SVM大坝安全性态预测模型,以k-CV验证误差最小作为优化目标,引入遗传算法对支持向量机的惩罚参数c和核函数参数g进行寻优。模型以影响因子作为输入,以效应量作为输出,采用训练样本对支持向量机进行训练,并使用训练好的模型预测效应量。根据概率统计理论中的3σ准则,建立大坝安全性态三级指标和判别准则。以某大型水库大坝为例,建立该大坝的GA-SVM模型,并与SVM模型和逐步回归模型进行了对比验证。预测结果表明,GA-SVM模型渗压预测值与实测值最接近,预测精度较SVM模型和逐步回归模型提高了约3倍。To make the prediction result of dam safety more accurate with support vector machine(SVM),a genetic algorithm-support vector machine(GA-SVM)dam safety prediction model was proposed based on the genetic algorithm optimization.With the minimum error of k-CV verification as the optimization target,the genetic algorithm was introduced to optimize the penalty parameter c and kernel function parameter g of SVM.The model took the influence factor as the input and the effect quantity as the output.Then the training sample data were used to train the support vector machine,and the trained model was used to predict the effect quantity.According to the 3σcriterion in the probability and statistics theory,the three-grade index and discriminant criterion of dam safety were established.Taking an actual large-reservoir dam as the case,the GA-SVM model of this dam was established and compared with the SVM model and stepwise regression model.The prediction results show that the predicted value of GA-SVM model is closest to the measured value,and its prediction accuracy is about three times higher than both SVM model and stepwise regression model.Therefore,the GA-SVM prediction model has a good prediction precision and can be used for the dam safety prediction.

关 键 词：水库大坝 安全性态 预测模型 遗传算法 支持向量机 k-折交叉验证 小波去躁 逐步回归 

分 类 号：TV698.1[水利工程—水利水电工程]