基于空间相关性的索力传感器优化布置及全桥索力反演预测研究
发布时间：2018-11-04 21:44
【摘要】：作为缆索体系桥梁中的重要承重结构,拉索成为桥梁结构安全评价的一个重要因素,索力监测在大跨径斜拉桥和悬索桥的健康监测系统中已经成为必不可少的监测项目。一个合理的拉索监测方案,特别是索力监测位置极大程度地影响着桥梁结构安全评估的准确性,也是建造经济节约型健康监测系统的关键。本文从拉索群荷载响应的空间相关性出发提出了基于索力相关性的传感器优化方法,并利用最优测点处的索力信息实现了对未监测位置处拉索索力的估计,达到了利用有限的传感器最大限度地获取全桥拉索荷载响应的目的。首先采用B样条拟合法提取由环境因素引起的索力趋势项为研究对象,以Pearson相关系数、最大信息系数(MIC)和互信息系数(MI)三个系数作为拉索群空间相关性的度量指标,深入地挖掘不同位置拉索间荷载响应的内在关联。Pearson相关系数只能描述变量间的线性关系,但是桥梁的荷载响应的复杂程度远远超过线性关联;最大信息系数因为噪声的存在使其难以发挥在相关性探索中的优势,而拉索间的散点图却呈现出“宽带”的特点;相比而言,基于核密度估计的互信息系数更能探索拉索群间的非线性关系,因此选取互信息系数作为拉索群空间相关性建模的依据;其次利用键能算法(BEA)对拉索群相关系数矩阵进行聚类分析,并根据测点在聚类关联度中的排列顺序进行传感器测点分类和最优测点选择。以南京长江三桥上游84根拉索为研究对象,以0.05为间距,讨论了相关性阈值取0.9~0.6时上游拉索群传感器优化布置方案。当阈值取0.9时,有接近1/2的拉索被选择为监测对象,并且最优测点的个数随着相关性阈值的减少而逐渐减少,优化结果论证了该方法的有效性;最后提出了基于粒子群算法的核极限学习机模型(PSO_KELM)以实现利用有限的监测拉索对未监测处拉索索力变化的估计。从预测精度、误差分布等角度对比分析了不同的激活函数和核函数下的极限学习机模型(ELM)、多元线性回归模型(MLR)和自适应回归样条模型(MARS)在全桥索力反演预测中的性能,发现具有RBF核函数的RBF_KELM模型具有更高的预测精度和泛化能力,索力预测最大均方根为2.79,并且绝对误差落在[-3,3]区间内的概率为99.54%,预测精度满足实际工程的需要;文中还利用MARS模型对基于空间相关性的传感器优化方法进行验证,验证了选取互信息系数为相关性指标的合理性和优化方法的有效性。
[Abstract]:As an important load-bearing structure in cable-system bridges, cable has become an important factor in the safety evaluation of bridge structures. Cable force monitoring has become an indispensable monitoring project in the health monitoring system of long-span cable-stayed bridges and suspension bridges. A reasonable cable monitoring scheme, especially the location of cable force monitoring greatly affects the accuracy of bridge structure safety assessment, and is also the key to the construction of economical and economical health monitoring system. Based on the spatial correlation of load response of cable group, a sensor optimization method based on cable force correlation is proposed in this paper, and the cable force estimation at unmonitored position is realized by using the cable force information at the optimal measuring point. The finite sensor is used to obtain the load response of the whole bridge cable to the maximum extent. Firstly, using B-spline fitting method to extract the trend term of cable force caused by environmental factors as the research object, taking Pearson correlation coefficient, maximum information coefficient (MIC) and mutual information coefficient (MI) as the measurement index of spatial correlation of cable group, the following three coefficients are used to measure the spatial correlation of cable group, such as Pearson correlation coefficient, maximum information coefficient (MIC) and mutual information coefficient (MI). The Pearson correlation coefficient can only describe the linear relationship between variables, but the complexity of load response of bridge is far more than linear correlation. Because of the existence of noise, the maximum information coefficient makes it difficult to play its advantage in the exploration of correlation, but the scattered plot between cables shows the characteristic of "broadband". In contrast, the mutual information coefficient based on kernel density estimation can better explore the nonlinear relationship between cable groups, so the mutual information coefficient is chosen as the basis of spatial correlation modeling of cable groups. Secondly, the correlation coefficient matrix of cable group is clustered by (BEA), and the sensor points are classified and the optimal points are selected according to the arrangement order of the measured points in the cluster correlation degree. Taking 84 cables in the upper reaches of Nanjing Yangtze River third Bridge as the research object and 0.05 as the spacing, the optimal arrangement scheme of the upstream cable group sensors is discussed when the correlation threshold is 0.90. 6. When the threshold is 0.9, nearly 1 / 2 of the cables are selected as the monitoring object, and the number of the optimal measuring points decreases with the decrease of the correlation threshold. The optimization results demonstrate the effectiveness of the method. Finally, a kernel limit learning machine model (PSO_KELM) based on particle swarm optimization algorithm is proposed to estimate the variation of cable forces in unmonitored cables using finite monitoring cables. This paper compares and analyzes the extreme learning machine model (ELM), under different activation function and kernel function from the angles of prediction precision and error distribution. The performance of multivariate linear regression model (MLR) and adaptive regression spline model (MARS) in full-bridge cable force inversion prediction is studied. It is found that the RBF_KELM model with RBF kernel function has higher prediction accuracy and generalization ability. The maximum root mean square (RMS) of cable force prediction is 2.79, and the probability of absolute error falling in the range of [-3] is 99.54. The prediction accuracy meets the needs of practical engineering. The MARS model is also used to validate the sensor optimization method based on spatial correlation, which verifies the rationality of selecting mutual information coefficient as the correlation index and the effectiveness of the optimization method.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U446
本文编号：2311229
[Abstract]:As an important load-bearing structure in cable-system bridges, cable has become an important factor in the safety evaluation of bridge structures. Cable force monitoring has become an indispensable monitoring project in the health monitoring system of long-span cable-stayed bridges and suspension bridges. A reasonable cable monitoring scheme, especially the location of cable force monitoring greatly affects the accuracy of bridge structure safety assessment, and is also the key to the construction of economical and economical health monitoring system. Based on the spatial correlation of load response of cable group, a sensor optimization method based on cable force correlation is proposed in this paper, and the cable force estimation at unmonitored position is realized by using the cable force information at the optimal measuring point. The finite sensor is used to obtain the load response of the whole bridge cable to the maximum extent. Firstly, using B-spline fitting method to extract the trend term of cable force caused by environmental factors as the research object, taking Pearson correlation coefficient, maximum information coefficient (MIC) and mutual information coefficient (MI) as the measurement index of spatial correlation of cable group, the following three coefficients are used to measure the spatial correlation of cable group, such as Pearson correlation coefficient, maximum information coefficient (MIC) and mutual information coefficient (MI). The Pearson correlation coefficient can only describe the linear relationship between variables, but the complexity of load response of bridge is far more than linear correlation. Because of the existence of noise, the maximum information coefficient makes it difficult to play its advantage in the exploration of correlation, but the scattered plot between cables shows the characteristic of "broadband". In contrast, the mutual information coefficient based on kernel density estimation can better explore the nonlinear relationship between cable groups, so the mutual information coefficient is chosen as the basis of spatial correlation modeling of cable groups. Secondly, the correlation coefficient matrix of cable group is clustered by (BEA), and the sensor points are classified and the optimal points are selected according to the arrangement order of the measured points in the cluster correlation degree. Taking 84 cables in the upper reaches of Nanjing Yangtze River third Bridge as the research object and 0.05 as the spacing, the optimal arrangement scheme of the upstream cable group sensors is discussed when the correlation threshold is 0.90. 6. When the threshold is 0.9, nearly 1 / 2 of the cables are selected as the monitoring object, and the number of the optimal measuring points decreases with the decrease of the correlation threshold. The optimization results demonstrate the effectiveness of the method. Finally, a kernel limit learning machine model (PSO_KELM) based on particle swarm optimization algorithm is proposed to estimate the variation of cable forces in unmonitored cables using finite monitoring cables. This paper compares and analyzes the extreme learning machine model (ELM), under different activation function and kernel function from the angles of prediction precision and error distribution. The performance of multivariate linear regression model (MLR) and adaptive regression spline model (MARS) in full-bridge cable force inversion prediction is studied. It is found that the RBF_KELM model with RBF kernel function has higher prediction accuracy and generalization ability. The maximum root mean square (RMS) of cable force prediction is 2.79, and the probability of absolute error falling in the range of [-3] is 99.54. The prediction accuracy meets the needs of practical engineering. The MARS model is also used to validate the sensor optimization method based on spatial correlation, which verifies the rationality of selecting mutual information coefficient as the correlation index and the effectiveness of the optimization method.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U446
【参考文献】
相关期刊论文 前10条
1 杨锡运;关文渊;刘玉奇;肖运启;;基于粒子群优化的核极限学习机模型的风电功率区间预测方法[J];中国电机工程学报;2015年S1期
2 朱银娟;李海波;;云制造环境下基于关联度的资源组合方法[J];计算机工程与应用;2016年05期
3 李武;;基于MIC的我国经济与R&D经费相关性研究[J];河北经贸大学学报;2015年03期
4 高凤;;互信息的最大信息系数法在股市关联度上的应用[J];新西部(理论版);2014年19期
5 张海洋;朱美琳;;基于MIC的持股集中度与股票价格关系研究[J];市场研究;2014年04期
6 刘念;张清鑫;李小芳;;基于核函数极限学习机的分布式光伏短期功率预测[J];农业工程学报;2014年04期
7 曾令男;丁建伟;赵炯;张力;刘英博;;基于互信息的复杂装备高维状态监测数据相关性发现与建模[J];计算机集成制造系统;2013年12期
8 余时伟;黄廷祝;刘晓云;陈武凡;;显著图引导下基于偏互信息的医学图像配准[J];仪器仪表学报;2013年06期
9 韩敏;梁志平;;改进型平均移位柱状图估算概率密度并对互信息作相关分析[J];控制理论与应用;2011年06期
10 王金亮;余海燕;胡松波;刘文君;;关于Neyman-Pearson基本引理的几个注记[J];数学杂志;2011年02期
相关硕士学位论文 前2条
1 崔娜;基于MIC方法的气候-疟疾敏感人群识别和脆弱区划研究[D];西安科技大学;2013年
2 邓彩凤;中文文本分类中互信息特征选择方法研究[D];西南大学;2011年
，本文编号：2311229
本文链接：https://www.wllwen.com/kejilunwen/daoluqiaoliang/2311229.html
