摘要
针对常规径向基函数(RBF)神经网络模型无法选择显著预报因子和易陷入局部最优解的问题,建立一种融合平均影响值(MIV)、改进果蝇算法(FOA)和RBF神经网络的大坝变形监测模型.通过引入MIV对水压、温度、时效三类预报因子进行筛选,并利用改进FOA算法获得RBF神经网络模型中最佳的spread值,以提高模型的稳定性和预报精度.为验证模型的有效性,以某混凝土重力坝位移监测数据为例,分别建立多元线性回归模型、常规RBF模型、MIV-RBF模型和MIV-改进RBF模型.研究结果表明MIV-改进RBF神经网络大坝变形监测模型预测稳定、精度高,预报效果好.
For the fact that the standard radial basis function（RBF）neural network can＇t select the significant factors and is easy to jump into local optimum,a fusion of the mean impact value（MIV）,improved fruit flies algorithm（FOA）and monitoring model of RBF neural network is established.Firstly,to select three kinds of forecast factor,that is,water pressure,temperature and prescription,the MIV is introduced.Then the FOA is used for searching the optimal spread value of RBF neural network model.By these two approaches,the stability and forecast precision of the model is improved.To verify the validity of the model,taking the displacement monitoring data of a concrete gravity dam for example,multiple linear regression model,the standard RBF model,the MIV-RBF model and MIV-improved RBF model are built respectively.The results show that the MIV-improved RBF neural network has great generalization ability,stable prediction and high precision.Meanwhile,the results of this monitor model of dam deformation prediction is remarkable and the model can be applied to practical engineering analysis of the deformation monitoring and early warning.
出处
《三峡大学学报（自然科学版）》
CAS
2016年第3期1-5,共5页
Journal of China Three Gorges University：Natural Sciences
基金
国家自然科学基金(51279050)
土石坝长效安全运行重大关键技术研究(201501033)
关键词
MIV算法
变量筛选
改进RBF神经网络
大坝变形监测模型
mean impact value（MIV）
variable selection
improved radial basis function（RBF）neural network
deformation monitoring model of dam
作者简介
通信作者：宁昕扬（1991-），男，硕士研究生，主要从事水工结构方面研究．E-mail：ninxiny@163．com
