高边坡时序位移滚动预测的SVM-Elman模型  

A Time Series Prediction Model of High Slope Displacement Based on Support Vector Machine and Elman Neural Network

作　　者：刘冲 沈振中 甘磊[1,2] 旦增赤列 严中奇[3] LIU Chong;SHEN Zhen-zhong;GAN Lei;DANZENG Chi-lie;YAN Zhong-qi(State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing 210098, China;College of Water Conservancy and Hydropower Engineering, Hohai University, Nanjing 210098, China;Tongxiang Water Conservancy Bureau, Tongxiang314500, China)

机构地区：[1]河海大学水文水资源与水利工程科学国家重点实验室,南京210098 [2]河海大学水利水电学院,南京210098 [3]浙江省桐乡市水利局,浙江桐乡314500

出　　处：《长江科学院院报》2019年第5期62-68,共7页Journal of Changjiang River Scientific Research Institute

基　　金：国家自然科学基金项目(51179062);2016年度江苏省普通高校学术学位研究生科研创新计划项目(KYZZ16_0284);江苏省自然科学基金青年基金项目(BK2012410);中央高校基本科研业务费项目(2014B11914)

摘　　要：基于支持向量机(SVM)和Elman神经网络,提出一种新的高边坡位移时序预测模型——SVM-Elman神经网络预测模型。在对实测数据学习的过程中,寻找最佳学习样本数和最佳测试样本数,利用经粒子群算法优化的SVM模型对边坡位移时间序列进行实时滚动预测;并运用Elman神经网络改进SVM的预测结果,得到SVM-Elman模型预测值,通过比较不同隐含层数的Elman神经网络对预测结果的影响,选择最佳隐含层数的SVM-Elman模型,实现对预测结果的改进。将SVM-Elman模型应用于某混凝土面板堆石坝左岸强卸荷岩体高边坡位移预测分析中,并与传统的SVM预测结果进行比较分析。结果表明,SVM-Elman模型在预测精度上有明显提高,预测结果科学可靠,在岩体高边坡时序位移预测中具有一定的工程应用价值。A new displacement time series predicting model was proposed by integrating support vector machine (SVM) and Elman neural network, named as SVM-Elman model. In the process of measured displacement data learning, by searching the best historical step and the best prediction step, SVM model was optimized by particle swarm algorithm to dynamically forecast the trend of development. In the meantime, Elman neural network has the ability of dynamically reflecting the development trend of the absolute error of SVM model prediction. By comparing the influence of different hidden layers of Elman neural network on the prediction results, the optimal number of hidden layer was determined for SVM-Elman model and hence modifying the predicted data of SVM in real time. The proposed SVM-Elman model was applied to the strong unloading high slope on the left bank of a concrete face rockfill dam, and the prediction result was compared with that of conventional SVM. Results demonstrate that the proposed model has superior accuracy and real application value in predicting the deformations of high slope.

关 键 词：边坡变形预测 支持向量机 ELMAN神经网络 SVM-Elman模型 粒子群优化算法 隐含层数 

分 类 号：TV698.1[水利工程—水利水电工程]