桥梁时变可靠度指标的改进粒子滤波预测算法

Improved Particle Filter Prediction Algorithm of TimeVariant Reliability Indices for Bridges

摘要:

基于健康监测时间序列数据，提出了桥梁动态可靠度指标的改进粒子滤波预测方法.首先，利用监测极值数据建立动态模型，将其作为粒子滤波算法的状态方程和监测方程；然后，采用贝叶斯动态线性模型(BDLM)为粒子滤波器提供随时间更新的动态建议分布，以解决传统粒子滤波算法的样本退化问题，同时增加了粒子滤波算法的鲁棒性及自适应性；进而利用改进的粒子滤波算法(IPF)，结合极值监测数据实现结构极值的动态预测，并结合一次二阶矩(FOSM)可靠性方法，实现桥梁结构可靠度指标的动态预测；最后通过在役桥梁工程实例与设计试验对所提模型和方法的合理性与有效性进行验证.

Abstract:

This paper proposes an improved particle filter (IPF) prediction approach of dynamic reliability indices for bridges based on monitoring time series data. First, the dynamic models, which can provide state equation and monitoring equation for the IPF, are built with the monitoring extreme data of bridges. Next, the Bayesian dynamic linear model (BDLM) is utilized to produce the realtime updated proposal distribution for IPF in order to solve the sample degradation problem and increase the robustness and adaptability of the traditional particle filter. After that, by using the IPF approach, the structural extreme information is dynamically predicted based on the monitoring extreme data, and dynamic reliability indices of bridges are predicted by using the first order second moment (FOSM) reliability method. Finally, three existing bridges and a designed experiment are provided to illustrate the feasibility and application of the proposed model and method.