基于抗扰观测与参数学习的混合动力车辆扭振特征识别
谢辉 , 郑涛 , 梁博强
天津大学学报(自然科学与工程技术版) ›› 2026, Vol. 59 ›› Issue (11) : 1153 -1165.
基于抗扰观测与参数学习的混合动力车辆扭振特征识别
Torsional Vibration Feature Identification of Hybrid Electric Vehicles Based on Disturbance Observation and Parameter Learning
为准确识别和分析混合动力汽车发动机启动过程中的扭转振动特性,提出了一种扭振的时频域特征识别算法.首先,基于传动系统轴系结构建立集中质量模型,分别在时域和频域对系统动态特性进行分析.在此基础上,引入扩张状态卡尔曼滤波器(ESKF)估计扭转减振器扭矩.其次,针对发动机启动过程中扭矩信号表现出的非平稳特性,采用连续小波变换(CWT)方法进行时频局部化分析,有效识别瞬态工况下的频域共振特征.进一步地,结合动力学模型与固有频率模型,提出了参数分区学习算法,以实现关键结构参数的自适应辨识,提升观测器在瞬态工况下的准确性与鲁棒性.仿真结果表明:在存在输入扰动与传感器噪声的情况下,ESKF 观测器的扭矩估计均方根误差(RMSE)为 4.15 N·m,相较于线性矩阵不等式(LMI)观测器的 9.11 N·m,估计精度提升了 54.45%.在连续小波变换分析中,在每倍频程频带数(VPO)为 15、采样频率为 50 Hz,采用 Morlet 小波识别出的共振频率为 14.698 Hz,与系统固有频率 14.53 Hz 的相对误差为 1.1%.此外,在结构参数存在 20%扰动的情况下,固定模型参数的观测器的 RMSE 为 18.96 N·m,而经参数学习算法校正后,RMSE 降至 4.23 N·m,表明所提方法提高了扭振特征在时频域内的识别精度.
To accurately identify and analyze the torsional vibration characteristics during the engine start-up process of a hybrid electric vehicle,a time-frequency domain feature identification algorithm for torsional vibration was proposed. First,a lumped mass model of the transmission shaft system was established,and the dynamic characteristics of the system were analyzed in both the time and frequency domains. On this basis,an extended state Kalman filter (ESKF)was adopted to estimate the torque of the torsional damper. Second,considering the nonstationary characteristics of the torque signal during engine start-up,the continuous wavelet transform(CWT)method was applied for time-frequency localization analysis and effectively identified the resonance features in the frequency domain under transient conditions. Furthermore,combined with the dynamic model and the natural frequency model,a parameter partition learning algorithm was proposed to achieve adaptive identification of key structural parameters,thereby improving the accuracy and robustness of an observer under transient operating conditions. Simulation results showed that,in the presence of input disturbances and sensor noise,the root mean square error(RMSE)of torque estimation by the ESKF observer was 4.15 N·m,representing an improvement of 54.45% compared with 9.11 N·m obtained by a linear matrix inequality(LMI)observer. In the CWT analysis,when the number of voices per octave was 15,and the sampling rate was 50 Hz,the resonance frequency identified by the Morlet wavelet was 14.698 Hz,with a relative error of 1.1% compared with the system natural frequency of 14.53 Hz. In addition,when the structural parameters had a 20% perturbation,the RMSE value obtained by a fixed-parameter observer was 18.96 N·m,while after correction using the parameter learning algorithm,it was reduced to 4.23 N·m. The results indicated that the proposed method improved the identification accuracy of torsional vibration features in the time-frequency domain.
| [1] |
|
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
王 博, 张振东, 于海生, |
| [14] |
|
| [15] |
|
| [16] |
|
| [17] |
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
|
| [22] |
蔡英凤, 窦 磊, 陈 龙, |
| [23] |
|
| [24] |
|
| [25] |
|
| [26] |
|
| [27] |
刘 辉, 张 伟, 张万年, |
| [28] |
|
| [29] |
|
国家自然科学基金资助项目(51906174)
/
| 〈 |
|
〉 |