大坝观测效应量的明确性和非明确性混合分量拟合模型

Hybrid definite and undefinite components fitting model for effect-size of dam observation

目前大坝观测效应量拟合模型可以分为两大类:明确性分量拟合模型和非明确性分量拟合模型。事实上,水压分量和温度分量采用明确性分量拟合模型是合适的,而时效分量具有非线性和非明确的特征,所以宜采用非明确性分量模型。鉴于此,提出明确性和非明确性混合分量拟合模型。采用能够较好地描述非线性变化的支持向量机(SVM)拟合时效分量,同时结合传统逐步回归分析,对于变化特征较为明显的水压分量与温度分量采用传统的统计模型进行描述。然后,将两大部分分量整合形成新的混合模型,并利用遗传算法GA对水压分量与温度分量进行修正。工程应用实例显示,该方法对实测资料的拟合比传统逐步回归方法更为精确,时效分量的总体变化趋势与逐步回归成果相似,但存在局部差异。

At present,the fitting model for effect-size of dam observation can be divided into two categories,i.e.definite component fitting model and undefinite component fitting model.Actually,the definite component fitting model is suitable for both the water pressure component and the temperature component,while the undefinite component fitting model is appropriate to be used for the time dependent component,as it has non-linear and definite features.On the basis of this,a hybrid definite and undefinite components fitting model is put forward herein.At first,the SVM(Support Vector Machine)which can better describe the non-linear change is adopted for fitting the time dependent component;meanwhile,the water pressure component and the temperature component with more obvious changing characteristics are described with the conventional statistical model in combination with the conventional stepwise regression analysis.After that,the two large parts of components are integrated into a new hybrid model,and then both the water pressure component and the temperature component are modified with the genetic algorithm GA.Actual engineering application case shows that the fitting of the measured data made with this method is more accurate than that from the stepwise regression method,from which the general changing trend of the time dependent component is similar to the result from the stepwise regression method,but local differences are still there.