A Novel Weighted Unscented Kalman Filter for Dynamic Load Identification

To address the limitations of traditional unscented Kalman filter (UKF)‐based algorithms—which typically require either additional displacement measurements or iterative optimization for load identification—this study proposes a fast and convenient load excitation identification algorithm. The method is developed based on a novel approach for calculating the weight factors in the unscented transformation. By appropriately adjusting these weight factors, the identification results align more closely with actual measurements, effectively mitigating the phenomenon of load drift. This enables the algorithm to operate using only acceleration measurements as observations. In addition, to facilitate rapid tuning, this study employs parameter sensitivity analysis to determine reasonable ranges for the covariance parameters and the key parameters governing the sigma point distribution. Numerical simulation results demonstrate that the proposed algorithm exhibits strong noise robustness, maintaining accurate load identification even under 20% measurement noise. The algorithm is also insensitive to initial state variables and shows strong adaptability to parameter variations. Furthermore, it is capable of accurately identifying the trend and effective values (e.g., peak and valley values) of the load even under incomplete acceleration measurement conditions. Shaking table test results further confirm that appropriately reducing the value of the measurement covariance matrix promotes algorithmic convergence while maintaining identification accuracy.