帷幕灌浆量区间预测的Bootstrap-IGWO-SVM模型研究  

Curtain grouting cement interval prediction using BootstrapIGWO-SVM model

作　　者：李凯[1] 任炳昱[1] 关涛[1] 余佳[1] 王佳俊[1] LI Kai;REN Bingyu;GUAN Tao;YU Jia;WANG Jiajun(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300350,China)

机构地区：[1]天津大学水利工程仿真与安全国家重点实验室,天津300350

出　　处：《水力发电学报》2022年第10期18-29,共12页Journal of Hydroelectric Engineering

基　　金：国家自然科学基金项目(51879186);中国长江三峡集团有限公司科研项目资助(BHT/0773)。

摘　　要：由于帷幕灌浆注灰量预测过程中存在地质参数、预测模型和输入数据的不确定性,传统的点预测结果存在误差,并且难以对不确定性进行量化。针对上述问题,本研究提出基于Bootstrap方法和改进灰狼算法的支持向量机(Bootstrap-IGWO-SVM)的帷幕灌浆量区间预测模型,量化了预测模型的不确定性。首先通过Bootstrap算法对初始训练集抽样生成样本数据集;其次,通过灰狼优化算法对惩罚因子C、RBF核函数方差g和损失因子p进行参数寻优,提高SVM算法的预测精度;再次,利用非线性收敛因子、动态权重因子、概率混沌图谱和Levy飞行对灰狼算法进行改进,解决灰狼算法局部搜索和全局搜索的平衡问题;最后,对构建的数据集分别使用IGWOSVM算法和随机森林方法分别预测得到系统误差和随机误差,并将两者累加得到总体误差,进而通过构建正态分布模型得到注灰量区间预测结果,实现了预测模型不确定性的量化。结果表明,改进的IGWO-SVM的预测精度为RMSE=85.32,R^(2)=0.53,MAE=45.64,相比GWO-SVM方法(RMSE=96.58,R^(2)=0.40,MAE=48.45)明显提升,相比BP神经网络算法(BPNN),极限学习机(ELM)存在明显精度优势;在置信度为99%下预测区间覆盖率(PICP)、预测区间宽度(MPIW)和宽度综合指标(CWC)分别为98.71%、363.59 kg/m、363.59 kg/m。Uncertainties exist in the geological parameters,prediction model,and input data of the curtain grouting cement predictions;the traditional point prediction suffers considerable errors and lacks a capability of quantifying the uncertainty level.This study describes a Bootstrap method and an improved grey wolf support vector machine(Bootstrap-IGWO-SVM),and develops an interval prediction model of curtain grouting cement with quantification of its uncertainty.First,a new data set is created from the initial training set sampling using the Bootstrap method;the grey wolf method is used to optimize the penalty factor C,RBF kernel function variance g,and loss factor p so as to improve prediction accuracy.Then,we improve the grey wolf method using the nonlinear convergence factor,dynamic weight factor,probabilistic chaos map,and Levy flight,so that the imbalance between local and global searchings can be eliminated.Finally,we estimate the system error and random error using the IGWO-SVM method and the random forest method respectively,and sum up them as the total error,thereby obtaining a cement interval prediction through building a normal distribution model.Thus,the uncertainty of prediction model is quantified.The results show the prediction accuracy of this improved IGWO-SVM is RMSE=85.3,R^(2)=0.53,and MAE=45.6,a significant improvement on the GWO-SVM’s values of 96.6,0.40and 48.5.It is also superior in prediction accuracy to the back propagation neural network(BPNN)or extreme learning machine(ELM).At the confidence level of 99%,its prediction interval coverage probability(PICP),mean prediction interval width(MPIW),and coverage width-based criterion(CWC)are 98.7%,363.6 kg/m,and 363.6 kg/m respectively.

关 键 词：灌浆量预测 改进的灰狼优化算法 支持向量机 区间预测 帷幕灌浆 

分 类 号：TV543[水利工程—水利水电工程]