基于改进复杂追踪算法的结构模态参数识别

Structural Modal Parameter Identification based on Improved Complex Pursuit Algorithm

以固定梯度的复杂追踪算法进行目标函数寻优时,收敛速度慢,易陷入局部极值,且存在针对不同的模型需人为选择合适的学习步长等不足,限制了该算法的实用性。为此,将最优步长思想引入复杂追踪算法,根据实时分离度自动调整步长,并根据分离信号的峭度值自适应地选择不同的非线性函数,以提高算法的计算精度,进而增强其实用性。为验证该改进复杂追踪算法识别结构模态参数识别的可行性与优越性,采用该方法分别识别了简支梁的数值模型和三层框架试验模型的模态参数,并与原方法进行识别结果对比,结果表明该改进算法可以较准确地识别结构模态参数。

Constant gradient based complex pursuit(CP) is prone to fall into local extremum, has slow convergence speed, and needs to artificially choose the appropriate learning step for different models. These shortcomings greatly limit the performance in practical application of this method. To solve the defects above, the idea of optimal step is introduced into the CP. In this idea, the step is automatically adjusted by the real-time degree of separation. At the same time, different nonlinear function can be chosen according to the kurtosis of the separated signal. Then, the computation accuracy of the CP algorithm is raised and its practical applicability is strengthened. To verify its practicability and reliability, the improved CP is applied to identify the modal parameters of a concrete simply-supported beam and an experimental three-story frame. The results are compared with those using the original CP algorithm. The accuracy of the improved CP in structural modal identification is demonstrated.