基于遗传算法的小波神经网络模型预测大坝变形  

Dam Deformation Prediction Using Wavelet Neural Network Model Based on Genetic Algorithm

作　　者：蒋园园[1,2] 卢献健[1,2] 郑中天 刘海锋[1,2] 

机构地区：[1]桂林理工大学测绘地理信息学院,广西桂林541004 [2]广西空间信息与测绘重点实验室,广西桂林541004

出　　处：《地理空间信息》2017年第7期99-101,114,共4页Geospatial Information

基　　金：国家自然科学基金资助项目(41461089);广西自然科学基金资助项目(2014GXNSFAA118288);广西"八桂学者"岗位专项经费资助项目;广西空间信息与测绘重点实验室基金资助项目(桂科能1207115-07;桂科能130511407)

摘　　要：为了提高大坝变形的预测精度,提出一种基于遗传算法的小波神经网络模型。首先通过对BP神经网络隐含层神经元的替换,弥补了网络易收敛于局部极小点的缺陷,增强了函数逼近能力,进而建立了小波神经网络大坝预测模型;再利用该模型对大坝变形训练集进行学习,并运用遗传算法选取全局最优参数。该方法充分利用了小波神经网络强大的非线性预测能力和遗传算法的全局优化搜索功能,弥补了BP神经网络存在的理论缺点。将其与小波神经网络、BP神经网络进行比较,实验结果表明该方法具有更优的局部预测值、更高的全局预测精度,适用于复杂的大坝变形预测。In order to improve the prediction accuracy of dam deformation, a new wavelet neural network model based on genetic algorithm was presented in this paper. Firstly, the substitute of the BP neural network hidden layer neurons was summed to weaken that network was easy to converge to local minimum points, and enhanced the function approximation ability. Secondly, the dam prediction of wavelet neural network model was established. Thirdly, the model was used to study the dam deformation training set, and genetic algorithm was used to select the optimal parameters. This method made full use of wavelet neural network's nonlinear predictive ability and genetic algorithm's global optimization searching, and avoided the defects of theoretical methods about the BP neural network. This method was compared with the wavelet neural network and BP neural network at last. The result proves that the new method can guarantee the optimum of local forecasts and have better overall prediction accuracy. It is feasible to apply the model in complicated forecasts.

关 键 词：大坝变形 小波神经网络 遗传算法 参数优化 

分 类 号：P258[天文地球—测绘科学与技术]