摘要
在渗流监控指标中,库水位之间、库水位与降雨之间存在严重的相关性.利用普通多元线性回归建立渗流监控模型中,监控指标之间存在的多重相关性影响参数估计,扩大模型误差,破坏模型的稳健性.为了克服多重相关性对模型的干扰,引入了能辨别系统信息与噪声的偏最小二乘回归,并编制了程序.算例分析表明,偏最小二乘回归模型所分离出的各个影响分量能对大坝实测变量的变化作出合理的物理成因解释,而且偏最小二乘回归模型的预测能力也远优于普通最小二乘回归模型,前者的预测误差平方和约只有后者的二十分之一.
Among the indicators of seepage monitoring model, there is serous collinearitiy between each water level, water levels and rainfalls. In a seepage monitoring model built by ordinary muhilinear regression, the muhicollinearity between each monitoring indicator will influence the parameter estimation, enlarge the model error and damage the robustness of model. To avoid multicollinearity's disturbance, partial least - squares regression which can identify system information and noise is introduced into the model, and a program is compiled. It is illustrated by a case that the components of partial least - squares model can give a reasonable physical interpretation to variation of prototype observation data, and predictive power of partial least - squares regression is stronger than ordinary mutilinear regression, the sum square predictive error of former is nearly one of twentieth of the latter.
出处
《郑州大学学报(工学版)》
CAS
2006年第2期117-119,123,共4页
Journal of Zhengzhou University(Engineering Science)
基金
河南省自然科学基金资助项目(511050100)
关键词
偏最小二乘回归
渗流监控模型
原型观测
partial least- squares regression
seepage monitoring model
prototype observation