基于改进无迹卡尔曼滤波器算法的滞回模型参数识别研究

Research on the identification of hysteresis model parameters based on the modified unscented Kalman filter algorithm

地震产生的周期荷载作用下,钢混桥墩结构表现出滞回行为。为描述滞回行为,研究者提出各类滞回模型,其中BWBN(Bouc-Wen-Baber-Noori)模型可以描述结构滞回行为的强度退化、刚度退化和捏拢效应等典型特征。此外,无迹卡尔曼滤波器UKF(unscented Kalman filter)算法是识别BWBN模型参数的高效方法,但当参数初始值与真实值的偏差过大及缺乏对系统的整体估计时,UKF算法识别过程受到局限。本文改进生成样本点规则,提出改进UKF算法。数值模拟结果表明,在无噪声条件下,改进UKF算法识别得到的参数估计值与准确值的误差平均为1.51%,最大误差为4%;在2%均方根RMS(root mean square)高斯白噪声条件下,误差平均为5.43%,最大误差为18%;在5%RMS高斯白噪声条件下,误差平均为8.9%,最大误差为26%和22%。改进UKF算法识别非线性滞回系统状态估计和BWBN模型参数更加准确和稳定。

The structure of reinforced concrete bridge pier exhibits hysteresis behavior when subject to cyclic loadings produced by earthquakes.To describe the hysteresis behavior,researchers have proposed various hysteresis models,among which BWBN(Bouc-Wen-Baber-Noori)model can describe typical characteristics including strength degradation,stiffness degradation and the pinching effect.Besides,UKF(unscented Kalman filter)algorithm is an efficient method to identify parameters of BWBN model but its identification process is restrictive when the deviation between the initial and real values of parameters is big enough and lacking of overall estimation for the system.In the paper,the rule of generating sample points is modified and the modified UKF algorithm is proposed.The numerical simulation results show that the average error between estimated and real value of parameters is 1.51%and the maximum error is 4%with no noise;the average error is 5.43%and the maximum error is 18%with 2%RMS(root mean square)Gaussian white noise;the average error is 8.9%and the maximum error is 26%or 22%with 5%RMS Gaussian white noise.The modified UKF algorithm is more accurate and robust to identify the state estimation of a non-linear hysteresis system and parameters of a BWBN model.