节点文献
大坝安全监测数据分析方法研究
Study on Analysis Methods for Dam Safety Monitoring Data
【作者】 李富强;
【导师】 刘国华;
【作者基本信息】 浙江大学 , 水工结构工程, 2012, 博士
【摘要】 大坝监测数据分析理论和方法的研究与应用已经取得了相当的进展,为保证大坝安全运行发挥了巨大的作用,但是,在数据分析方面依然存在许多问题和不足。针对现有分析方法和分析模型中存在的问题和不足,本文以混凝土坝变形监测数据分析为主,将其它领域的研究理论和分析方法应用到大坝监测数据的分析中,致力于提高监测数据分析时模型的预测精度,更加有效合理地实现对大坝运行现状的评价,满足实际工程应用的需要。为了避免回归模型可能存在的伪回归现象。本文利用协整理论检验大坝监测变量及相关环境影响因子数据序列的平稳性,对于存在协整关系的时间序列,采用误差修正模型来描述变量之间的长期均衡和短期非均衡关系,以提高模型的拟合精度和预测能力。为了评价大坝运行中坝体的安全状态和结构性态,根据平稳系统自回归模型特征多项式的根距离单位圆的远近,在一定程度上反映了该系统平稳性的变化情况,本文据此提出一种安全监控指标。时间序列的高阶统计量包含了二阶统计量所没有的大量丰富信息,能更好地反映系统的性态。本文介绍了现代谱估计及双谱估计理论、原理及方法,通过钢筋混凝土梁损伤试验验证了监测数据的双谱能较好地反映结构性态的变化,并尝试用于大坝变形监测数据的分析来评价大坝的结构性态变化趋势。时效分析在大坝变形监测中具有十分重要的意义,本文假定大坝系统为时不变系统,将时效作为反映大坝结构性态的状态变量,采用EM算法,利用状态空间模型进行时效分量的估计,实例分析验证了该方法不但具有较好的拟合及预测能力,而且可以有效提取出时效分量用于评价大坝的运行性态。自变量的多重共线性及随机噪声干扰往往会造成回归模型出现过拟合现象,使得模型拟合精度很高,但是预测能力很差,不能有效地用于大坝安全监控的预测预警。本文应用自组织数据挖掘技术的数据分组处理(GMDH)算法建立分析模型,增强模型稳健性,提高模型的预测能力,实例分析验证了该方法的有效性。
【Abstract】 Analysis theory and Methods for dam safety monitoring data, playing a major role in guaranteeing the dam safety, have been made considerable progress. However, there are still many problems and and deficiencies in the data analysis. In this paper, research theory and analytical methods in the other fields are introduced and applied to the analysis of dam monitoring data such as the analysis of deformation monitoring data of concrete dam. The main purpose of study in this paper is to solve some problems and to correct some deficiencies of the analysis of dam monitoring data in the existing, for improving the monitoring data analysis precision of the model and achieving more effective and rational assessment of the state of the dam, in order to meet the needs of practical engineering applications.In order to avoid possible suprious regression in regression model, in this paper, cointegration theorom is used to test whether a set of data of dam monitoring variables are stationary time series. The cointegrated series can be represented by error correction models to describe long run equilibrium and short run unequilibrium among them, in order to improve the model fittness and predictions capacity.In order to evaluate the security state and the structural change trend of the dam, a kind of safety monitoring index is introduced, according to the distance to the unit circle of roots which are calculated from the characteristic polynomial of stationary autoregression system reflects the properties of stationary system.As the higher-order statistics of time series which contains more information than a second-order statistics can better reflect the properties of system, this paper describes the modern spectral estimation and bispectrum estimation theory, principles and methods, the analysis result of a damage experiment of two reinfoced concrete beams indicates bispectrum can effectively reflect the change of structure status, and try for the analysis of dam monitoring data to assess the structural state of the dam.Considering the importance of time-effect displacement in deformation monitoring data analysis, suppose that the dam system is a time invariable system, the time-effect variable is selected as state variable to discribe the structural properties of dam body. The parameters of state space model arc estimated by using the EM algorithm. The example analysis result shows that the state space model has good precision of fitness and forecastion, and time-effect variable can effectively extract from the monitoring data of dam used to evalute the dam status.Multicollinearity of independent variables and random noise tend to result in overfitting of the regression model, making the model fitting accuracy is high, but the predictive ability is poor, so that the regression model can not be effectively applied to forecast for dam safety monitoring. In this paper, the group method of data handling(GMDH) of self-organizing data mining is used to make model in order to enhance the robustness of the model and to improve the model’s predictive ability, example analysis result validates the effectiveness of the method.
【Key words】 Dam; Time Series; Cointegration Test; Error Correction Model; Higher-order Statistics; State Space Model; GMDH Method;