摘要
水库裂缝开合情况对于水库的安全运行极为重要。将主成分分析法与广义回归神经网络结合在一起,进行水库裂缝开合度的预测。结果表明:应用主成分分析与广义回归神经网络相耦合的模型可以很好的反映环境因子(水压力因子、温度因子、时效因子)与水库裂缝开合度之间的非线性函数映射关系。同时利用Matlab软件对新疆某寒区水库裂缝的开合度进行了实例分析和预测。预测结果显示,水库裂缝开合度的最大相对误差分别8.14%,相关性系数为0.984 7,具有较高的预报精度。通过主成分分析与广义回归神经网络相耦合的方法,有效的消除了原指标间的相关性,降低了神经网络的输入,提取了对因变量解释性最强的成分,使广义回归神经网络的输入层节点数由原来的8个减少到2个,起到了网络结构的简化,增强了网络的稳定性。耦合模型弥补了最小二乘回归无法有效识别和消除因子间多重相关性影响的不足,为水库裂缝开合度、大坝位移等指标预测提供了新的思路和方法。
Opening and closing of reservoir cracks plays an important role to the safety of reservoir.Although the principal component analysis can effectively deal with the problems of multi-collinearity and non-linearity among variables.The method of neural network is an ideal tool to deal with the problem of non-linearity,but serious correlation of input data will make the network unsteady.An attempt has been made to investigate the possibility of using generalized regression neural network to predict the opening-closing degrees of reservoir cracks in this paper.Results show that the principal component analysis and the generalized regression neural network model can be used to reflect well the non-linear function reflection relation of environmental factors such as water pressure(KH),temperature(KT) and time effect(Kθ) with opening-closing degrees of reservoir cracks.The combination method of the principal component analysis and the generalized regression neural network was used to predict opening-closing of reservoir cracks by using Matlab software in cold area in Xinjiang.The results of prediction indicate that the maximum relative error and correlation coefficient are 4.87% and 0.944 2,respectively,and representing a high precision of prediction.Therefore the method of combination is better than that of single one,which eliminates the correlation index of samples and reduces the input dimension of neural network,and extracts the principal components with optimum explanation to dependent variables.It makes the node numbers of generalized regression neural network input layer cut down from eight to two,simplifies the structure and strengthens the stability of neural network.The combination method indicates that the proposed model is more reliable and has a better performance than conventional methods.In contrast to the least square regression method,the combination model can effectively identify and eliminate the effects caused due to the factors' multi-interrelation.The combination model provides a newly effective and feasible way for forecasting the indicators of the opening-closing degrees of reservoir cracks and displacement analysis of dams.
出处
《干旱区地理》
CSCD
北大核心
2011年第4期584-590,共7页
Arid Land Geography
基金
国家自然科学基金项目(41071026)
国家重点基础研究发展计划973项目(2009CB421302)资助
关键词
主成分分析
广义回归神经网络
水库裂缝开合度
预测模型
principal component analysis
GRNN
opening-closing degrees of reservoir cracks
prediction model.
