时间序列预测法在岩土变形问题中的应用研究

发布时间：2018-08-18 14:55

【摘要】：岩土工程变形是工程系统内部复杂力学机制的宏观反应,蕴含了施工过程中的力学演化信息,若能从中挖掘演化规律,利用已有的实测变形数据建模预测未来变形量,进而反馈于原设计,及时调整施工方案或采取相应处理措施,可有效降低发生工程事故的可能性。该方法成功避开了复杂的岩土变形机理,可作为工程信息化施工和动态控制的有效途径。因此,针对岩土工程的变形预测和控制研究具有重要意义。岩土工程多以岩土体作为工程环境或工程材料,岩土体是一种非均质的、各向异性的弹塑黏性体,加之地质条件的复杂性,使其力学参数和力学现象都具有很强的随机性和不确定性,导致岩土工程的变形预测和控制具有相当的难度。此外,岩土工程变形还受到工程地质条件、场地环境条件、地面荷载、施工方法、施工进度、时间和温度等多种因素影响,使其变形序列除了具有岩土力学变化的内在规律外,通常还带有一定的随机性,即可将岩土实测变形序列分解为趋势序列和随机序列。其中,趋势序列体现了岩土工程变形的内在规律,是变形预测的主要依据；随机序列属于噪声序列,具有一定的平稳性,若选择人为剔除该部分信息,会降低预测结果的精度和真实性。因此,在岩土变形预测过程中,应针对趋势序列和随机序列的各自特征分别建立预测模型进行分析。本文基于时间序列预测法理论,以岩土工程变形实测数据为基础数据,结合小波变换、粒子群算法优化的最小二乘支持向量机(PSO-LSSVM)和自回归移动平均模型(ARMA)提出了联合的岩土变形预测方法和模型,基本思路是：对于施工前期的变形实测数据,首先利用Db4正交小波将其分解为趋势时间序列和随机时间序列；然后,针对趋势时间序列,先采用相空间重构技术进行预处理,再建立PSO-LSSVM模型对其进行预测,针对随机时间序列,直接利用EViews软件中的ARMA模型对其进行预测；最后将两个子序列的预测值叠加作为最终预测结果。将本文方法分别用于基坑工程实例和地基工程实例的变形预测分析,充分验证了预测方法和模型的有效性。
[Abstract]:Geotechnical engineering deformation is the macroscopic response of the complex mechanical mechanism in the engineering system, which contains the mechanical evolution information in the construction process. If the evolution law can be excavated, the future deformation can be predicted by using the existing measured deformation data modeling, and then feedback to the original design, timely adjusting the construction scheme or taking corresponding treatment measures, it can be effective. This method successfully avoids the complicated deformation mechanism of rock and soil and can be used as an effective way for information construction and dynamic control of engineering.
Geotechnical engineering mostly uses geotechnical body as engineering environment or material. Geotechnical body is a kind of heterogeneous and anisotropic elastic-plastic viscous body. In addition, the complexity of geological conditions makes its mechanical parameters and mechanical phenomena have strong randomness and uncertainty, which makes it difficult to predict and control the deformation of geotechnical engineering. In addition, the deformation of geotechnical engineering is also affected by many factors, such as engineering geological conditions, site environment conditions, ground load, construction method, construction schedule, time and temperature, which make the deformation sequence not only have the inherent law of geomechanical changes, but also have a certain randomness, that is, the measured deformation sequence can be decomposed into trend sequence. The trend sequence reflects the inherent law of geotechnical engineering deformation and is the main basis of deformation prediction; the random sequence belongs to noise sequence and has a certain degree of stability. If this part of information is removed artificially, the accuracy and authenticity of prediction results will be reduced. The prediction model is established by analyzing the respective characteristics of trend series and random sequences.
In this paper, based on the theory of time series prediction method, a combined prediction method and model of geotechnical deformation is proposed, which is based on the measured data of geotechnical engineering deformation, combined with wavelet transform, least squares support vector machine (PSO-LSSVM) optimized by particle swarm optimization and autoregressive moving average model (ARMA). The basic idea is as follows: For the pre-construction period, the prediction method and model are combined. First, Db4 orthogonal wavelet is used to decompose the deformation data into trend time series and random time series. Then, for trend time series, phase space reconstruction technique is used to pre-process the deformation data, and then PSO-LSSVM model is established to predict the deformation data. For the random time series, the ARMA model in EViews software is directly used to advance the deformation data. Finally, the prediction values of the two subsequences are superimposed as the final prediction results. The proposed method is applied to the deformation prediction analysis of foundation pit engineering and foundation engineering respectively, which fully verifies the effectiveness of the prediction method and model.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU43

【参考文献】