大型建筑物沉降变形的动态预测方法研究

Dynamic Prediction Methods of Settlement and Deformation of Large Buildings

目的精准预测建筑物沉降的规律及建筑物的变形。方法小波神经网络具有良好的时频局域化性质和神经网络的自学习功能。通过小波分解进行平移和伸缩变换后得到的级数,具有小波分解的一般逼近函数的性质,因此可以用来预报。回归分析的方法可以定量地分析出变型设计过程中设计变量与性能指标之间相互依赖的不确定关系,以此揭示出产品性能指标与影响其值变化的设计变量之间的内在关系。对回归分析模型和小波神经网络模型做简单介绍,以宿州市某建筑大楼的沉降点观测数据为例,对2种模型的预测结果进行检验,在变形监测中分析其精度和可行性。结果回归分析模型预测误差最大值为-0.4 mm,最小值为0.1 mm;小波神经网络模型预测误差最大值为-0.21 mm,最小值为-0.01 mm。结论通过实例证明了2种模型的可行性,为变形分析中将影响变形的直接因素纳入模型提供了一定的参考。由于不同建筑物的荷载情况等因素的差异,模型的运用可能有一定的局限性,仍需要大量的实例进行验证,在有些情况下需要将荷载因子进行变换,才能获得较好的拟合度。

Objective To accurately predict the law of building settlement and deformation prediction of buildings. Methods Wavelet neural network has good time-frequency localization and self-learning function of neural network. It was used to predict the properties of the series obtained by the wavelet decomposition of the series and the transformation of the series,which had the general function of the wavelet decomposition. By means of regression analysis,the relationship between the design variables and the performance indexes of the variant design process were analyzed quantitatively,and the intrinsic relationship between the product performance indexes and the design variable affecting its value were revealed. In this paper,the regression analysis model and wavelet neural network analysis were briefly introduced. And then with the observation data of the settlement point of a building block in Suzhou city as an example,the prediction results of the two models were tested,and the accuracy and feasibility of the deformation monitoring were analyzed. Results The maximum value of regression analysis model prediction error is-0. 4 mm,and the minimum value is 0. 1 mm; the maximum value wavelet neural network model prediction error is-0. 21 mm and the minimum value is-0. 01 mm. Conclusion The feasibility of the two models was proved by an example,which provides a reference for the direct factors affecting the deformation in the deformation analysis. At the same time,because of the difference of buildings for different load conditions and other factors,this model may have certain limitations,which still need a large number of examples to verify. In some cases,it needs to transform the load factor to obtain better fitting degree.