小波人工神经网络在建筑沉降预测中的应用研究

发布时间：2018-09-07 19:34

【摘要】：随着我国经济的快速发展及城市化水平的不断提高,城市可利用的土地资源正不断减少,各类高层建筑正迅速崛起。由于楼层的增加,荷载的增加,其施工将给建筑物本身及周边建筑群体带来复杂的形变影响。其中,最常见的是导致其发生不均匀沉降,若沉降严重则将危及建筑物的安全。变形监测作为信息化施工的关键环节,贯穿于建筑物设计期、施工期和运营期的整个过程,工程参建各方都对监测工作和数据分析给予了极大的重视。近年来,为探索出一种快速有效的沉降预测的方法,许多学者从理论与实践等多方面进行了大量的探索与研究,并取得了一定的成效,但也存在着许多的问题与不足。本文根据建筑地基沉降的特点,以及目前在该领域所广泛研讨的热点方法,将具有自学习、自组织且非线性逼近能力较好的人工神经网络模型纳入建筑沉降的预测中来,以BP神经网络为基础,并利用小波分析等方法对传统的网络模型进行了优化改进。通过实例工程的变形预测对传统网络模型与改进模型进行了分析与研究,并对其预测效果进行了评价,结果比较理想。从而表明小波分析与神经网络模型结合在建筑沉降预测中是可行的,且具有广阔的工程应用价值。本文主要从以下几个方面作了研究： (1)研究了BP神经网络算法。对单一的BP神经网络模型算法的局限性进行分析,针对传统网络模型存在的问题,对其进行了优化改进,较好克服了易形成局部极小而得不到全局最优、训学习效率低、收敛速度慢等问题,并将改进模型应用于变形预测。 (2)对小波分析进行研究。结合MATLAB软件探讨了小波分析在信号去噪领域中的应用,研究了利用小波分析实现信号去噪的方法,以及小波函数选取、阈值选取和小波分解、重构等问题,合理地运用小波分析对变形监测数据进行去噪预处理,以求预测结果更加准确。 (3)探讨了小波分析和神经网络模型的结合方式。二者的结合通常有两种式：一种是辅助式结合,也称为松散型结合方式；另一种是嵌入式结合,也即紧致型结合方式。 (4)以BP神经网络模型为基础,借助MATLAB,将改进的BP神经网络、辅助式小波神经网络和嵌入式小波神经网络模型应用于实际工程的沉降预测当中,通过和实测值的对比,分析比较三种模型的整体性能。结果表明,后两种小波神经网络的组合模型精度大体相当,预测效果明显优于单一的BP神经网络模型。最后对本文的不足之处作了简要的说明。
[Abstract]:With the rapid development of economy and the improvement of urbanization level, the available land resources in cities are decreasing, and various kinds of high-rise buildings are rising rapidly. Because of the increase of floor and the increase of load, the construction will bring complex deformation effect to the building itself and the surrounding buildings. Among them, the most common is to cause uneven settlement, if the settlement will endanger the safety of the building. Deformation monitoring, as a key link of information construction, runs through the whole process of building design period, construction period and operation period. All parties involved in the project pay great attention to monitoring work and data analysis. In recent years, in order to explore a rapid and effective method of settlement prediction, many scholars have made a great deal of exploration and research in theory and practice, and achieved certain results, but there are also many problems and shortcomings. In this paper, according to the characteristics of building foundation settlement and the hot methods which are widely studied in this field, the artificial neural network model with self-learning, self-organization and better nonlinear approximation ability is applied to the prediction of building settlement. Based on BP neural network and wavelet analysis, the traditional network model is optimized and improved. The traditional network model and the improved model are analyzed and studied through the deformation prediction of practical engineering, and the prediction effect is evaluated. The results are satisfactory. It shows that the combination of wavelet analysis and neural network model is feasible in building settlement prediction and has broad engineering application value. This paper mainly studies the following aspects: (1) the BP neural network algorithm is studied. The limitation of the single BP neural network model algorithm is analyzed. Aiming at the problems existing in the traditional network model, the optimization and improvement are carried out to overcome the local minima easily formed but not the global optimum, and the training and learning efficiency is low. The improved model is applied to the deformation prediction. (2) the wavelet analysis is studied. This paper discusses the application of wavelet analysis in signal denoising with MATLAB software, studies the method of signal denoising using wavelet analysis, and the selection of wavelet function, threshold selection, wavelet decomposition, reconstruction and so on. In order to obtain more accurate prediction results, wavelet analysis is used to preprocess the deformation monitoring data reasonably. (3) the combination of wavelet analysis and neural network model is discussed. There are usually two types of combination: one is auxiliary combination, also known as loose combination; the other is embedded combination, that is, compact combination. (4) based on BP neural network model, The improved BP neural network, the auxiliary wavelet neural network and the embedded wavelet neural network model are applied to the settlement prediction of practical engineering with the help of MATLAB,. The overall performance of the three models is analyzed and compared with the measured values. The results show that the combined models of the latter two kinds of wavelet neural networks have similar accuracy and the prediction effect is obviously better than that of the single BP neural network model. At last, the deficiency of this paper is briefly explained.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP183;TU196.2

【参考文献】