基于Gibbs抽样的结构时域载荷识别

Structural dynamic load identification in time domain based on Gibbs sampling

载荷识别的病态问题往往采用正则化技术处理,不过传统正则化方法所选取的正则化参数是恒定不变的,导致识别出的载荷精度不是很高。提出了基于Gibbs抽样的结构时域载荷识别方法,将未知载荷和测量噪声假设为随机变量,建立了载荷识别的多层贝叶斯模型,采用Gibbs抽样法获得载荷的后验值。相比于传统正则化方法,该法具有本征的自适应正则化性能。数值结果表明,该法可提高载荷识别精度,自适应的正则化参数具有良好的优越性。

The regularization technique is often employed to deal with the ill-posed problem of load identification.However,regularization parameters obtained with the traditional regularization technique are constant. Here,a novel approach of load identification in time domain based on Gibbs sampling method was proposed to assume unknown loads and measurement noise to be stochastic variables. The hierarchical Bayesian model of load identification was built. Gibbs sampling was adopted to obtain the posterior probability density distributions of the identified loads. Numerical simulations were performed to demonstrate the effectivenessof the proposed method by comparing the simulated results using this method with those using Tikhonov regularization method based on L-curve and GCV criterion.