混凝土重力坝变形监测统计模型研究
发布时间：2018-11-04 19:47
【摘要】：随着一些大坝、高坝的修建，水电在我国能源结构中发挥着越来越重要的作用。同时，大坝的工作性态、诊断大坝的健康状况也就成为坝工工作人员的重中之重。大坝所处环境比较复杂，其在运行过程中可能会发生一些未知的不利因素，所以必须对大坝进行安全监测，一旦发现异常，及时分析并找出原因，，就可为大坝的安全补救赢得时间。目前在大坝监测数据采集上已经比较先进。但是对数据样本的建模分析还处于半理论、半经验阶段，为此监测数据分析就变得比较棘手。变形监测是大坝安全监测的最主要项目，也是能够直观反映大坝安全与否的最可靠监测量，而统计模型因其简单、直观而被广大监测人员所使用。因此，本文主要以统计分析法对大坝变形监测资料作了以下分析研究： (1)简单介绍了本文大坝变形监测采集数据所用到的视准线法和前方交会法。在参考前人研究成果的基础上首先分析了组成大坝变形监测模型的基本自变量因子，包括水压分量、温度分量和时效分量。接着讨论了目前比较常用的统计分析方法——多元回归分析法，多元回归分析的缺点是没有考虑自变量之间的多重共线性问题，为此将多元回归分析法进行优化得到逐步回归分析法，对工程实例进行逐步回归分析得到比较满意的结果。 (2)逐步回归分析提高大坝监测模型质量的代价是舍弃严重线性相关的自变量项，这在大坝自变量因子对位移量的解释上就不是特别理想。偏最小二乘法有效避免了普通多元回归和逐步回归分析的缺点。它的基本原理是通过将自变量的高维数据空间投影到相应的低维特征空间，得到相互正交的特征向量，且在选取特征向量时强调自变量对因变量的解释和预测，再建立自变量和因变量的特征向量间的一元线性回归关系。既避免了共线性问题，同时也考虑了自变量对因变量的解释。最后通过工程实例验证了偏最小二乘法在大坝变形监测模型上的预测稳定性。 (3) BP神经网络是由大量神经元按照一定的拓扑结构生成的非线性动态系统，其训练时对初始权值和阈值的随机选取往往使网络不稳定或者陷入局部极小值。遗传算法通过选择、交叉、变异三种方式的操作，保持模型稳定并不断产生新的变异，具有很好的全局搜索能力。将遗传算法应用于BP网络进行初始权值和阈值的优化，建立GA-BP模型，通过工程实例验证，GA-BP模型要比单纯的BP神经网络模型有更好的训练精度和稳定性。 (4)针对传统大坝安全监测单测点建模只能反映大坝局部的变形情况。本文引入测点坐标分析了大坝监测模型的多维多测点多方向模型，通过偏最小二乘法确定模型回归系数，分析了环境量与大坝变形量之间的关系，实例分析结果表明多维多测点模型具有更好的概括性和全面性。
[Abstract]:With the construction of some dams and high dams, hydropower plays a more and more important role in the energy structure of our country. At the same time, the working state of the dam, the diagnosis of the dam's health condition becomes the most important for the dam workers. The dam is in a complex environment and some unknown unfavorable factors may occur during its operation, so it is necessary to monitor the safety of the dam. Once the anomalies are found, the causes should be analyzed and found in time. It will buy time for the safety remedy of the dam. At present, the dam monitoring data collection has been relatively advanced. However, modeling and analysis of data samples is still in the semi-theoretical and semi-empirical stage, so monitoring data analysis becomes more difficult. Deformation monitoring is the most important project of dam safety monitoring, and it is also the most reliable monitoring quantity which can directly reflect dam safety or not, and the statistical model is used by the vast number of monitors because of its simplicity and intuitionism. Therefore, this paper mainly uses the statistical analysis method to analyze the dam deformation monitoring data as follows: (1) this paper simply introduces the collimation method and the forward intersection method used in the dam deformation monitoring data collection. On the basis of previous research results, the basic independent variable factors, including water pressure component, temperature component and aging component, which constitute the dam deformation monitoring model, are first analyzed. Then it discusses the multivariate regression analysis method, which is commonly used at present. The shortcoming of multivariate regression analysis is that it does not consider the problem of multiple collinearity between independent variables. For this reason, the method of multiple regression analysis is optimized to obtain the stepwise regression analysis method, and the results of stepwise regression analysis of engineering examples are satisfactory. (2) the cost of improving the quality of dam monitoring model by stepwise regression analysis is to abandon the seriously linear dependent independent variables, which is not particularly ideal in the interpretation of the displacement by the independent variables of the dam. The partial least square method effectively avoids the shortcomings of ordinary multivariate regression and stepwise regression. Its basic principle is to project the high-dimensional data space of the independent variable to the corresponding low-dimensional feature space, and to obtain the orthogonal eigenvector, and to emphasize the interpretation and prediction of the dependent variable by the independent variable in the selection of the eigenvector. Then the linear regression relationship between the eigenvector of independent variable and dependent variable is established. Not only the problem of collinearity is avoided, but also the interpretation of dependent variables by independent variables is considered. Finally, the prediction stability of partial least square method in dam deformation monitoring model is verified by an engineering example. (3) BP neural network is a nonlinear dynamic system generated by a large number of neurons according to a certain topological structure. The random selection of initial weights and thresholds often makes the network unstable or fall into local minima. Genetic algorithm (GA) can keep the stability of the model and produce new mutation through the operation of selection, crossover and mutation, so it has a good global search ability. Genetic algorithm (GA) is applied to optimize the initial weights and thresholds of BP network, and the GA-BP model is established. It is verified by an engineering example that the GA-BP model has better training accuracy and stability than the simple BP neural network model. (4) the single point modeling for traditional dam safety monitoring can only reflect the local deformation of the dam. In this paper, the multi-dimensional multi-point multi-direction model of dam monitoring model is analyzed by means of measuring point coordinates. The regression coefficient of the model is determined by partial least square method, and the relationship between environmental quantity and dam deformation is analyzed. The analysis results show that the multi-dimensional multi-point model has better generality and comprehensiveness.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TV698.11
本文编号：2310962
[Abstract]:With the construction of some dams and high dams, hydropower plays a more and more important role in the energy structure of our country. At the same time, the working state of the dam, the diagnosis of the dam's health condition becomes the most important for the dam workers. The dam is in a complex environment and some unknown unfavorable factors may occur during its operation, so it is necessary to monitor the safety of the dam. Once the anomalies are found, the causes should be analyzed and found in time. It will buy time for the safety remedy of the dam. At present, the dam monitoring data collection has been relatively advanced. However, modeling and analysis of data samples is still in the semi-theoretical and semi-empirical stage, so monitoring data analysis becomes more difficult. Deformation monitoring is the most important project of dam safety monitoring, and it is also the most reliable monitoring quantity which can directly reflect dam safety or not, and the statistical model is used by the vast number of monitors because of its simplicity and intuitionism. Therefore, this paper mainly uses the statistical analysis method to analyze the dam deformation monitoring data as follows: (1) this paper simply introduces the collimation method and the forward intersection method used in the dam deformation monitoring data collection. On the basis of previous research results, the basic independent variable factors, including water pressure component, temperature component and aging component, which constitute the dam deformation monitoring model, are first analyzed. Then it discusses the multivariate regression analysis method, which is commonly used at present. The shortcoming of multivariate regression analysis is that it does not consider the problem of multiple collinearity between independent variables. For this reason, the method of multiple regression analysis is optimized to obtain the stepwise regression analysis method, and the results of stepwise regression analysis of engineering examples are satisfactory. (2) the cost of improving the quality of dam monitoring model by stepwise regression analysis is to abandon the seriously linear dependent independent variables, which is not particularly ideal in the interpretation of the displacement by the independent variables of the dam. The partial least square method effectively avoids the shortcomings of ordinary multivariate regression and stepwise regression. Its basic principle is to project the high-dimensional data space of the independent variable to the corresponding low-dimensional feature space, and to obtain the orthogonal eigenvector, and to emphasize the interpretation and prediction of the dependent variable by the independent variable in the selection of the eigenvector. Then the linear regression relationship between the eigenvector of independent variable and dependent variable is established. Not only the problem of collinearity is avoided, but also the interpretation of dependent variables by independent variables is considered. Finally, the prediction stability of partial least square method in dam deformation monitoring model is verified by an engineering example. (3) BP neural network is a nonlinear dynamic system generated by a large number of neurons according to a certain topological structure. The random selection of initial weights and thresholds often makes the network unstable or fall into local minima. Genetic algorithm (GA) can keep the stability of the model and produce new mutation through the operation of selection, crossover and mutation, so it has a good global search ability. Genetic algorithm (GA) is applied to optimize the initial weights and thresholds of BP network, and the GA-BP model is established. It is verified by an engineering example that the GA-BP model has better training accuracy and stability than the simple BP neural network model. (4) the single point modeling for traditional dam safety monitoring can only reflect the local deformation of the dam. In this paper, the multi-dimensional multi-point multi-direction model of dam monitoring model is analyzed by means of measuring point coordinates. The regression coefficient of the model is determined by partial least square method, and the relationship between environmental quantity and dam deformation is analyzed. The analysis results show that the multi-dimensional multi-point model has better generality and comprehensiveness.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TV698.11
【参考文献】
相关期刊论文 前1条
1 杨杰,吴中如,顾冲时;大坝变形监测的BP网络模型与预报研究[J];西安理工大学学报;2001年01期
本文编号：2310962
本文链接：https://www.wllwen.com/kejilunwen/shuiwenshuili/2310962.html
