Abstract:The performance of disturbance observers is strongly influenced by the level of prior knowledge about the disturbance model. The simultaneous input and state estimation (SISE) algorithm is widely recognized for providing unbiased minimum-variance estimates under arbitrary disturbance models. In contrast, the Kalman filter-based disturbance observer (KF-DOB) achieves minimum mean-square error estimation when the disturbance model is fully specified. However, practical scenarios often fall between these extremes, where only partial knowledge of the disturbance model is available. This paper investigates the inherent bias-variance trade-off in KF-DOB when the disturbance model is incomplete. We further show that SISE can be interpreted as a special case of KF-DOB, where the disturbance noise covariance tends to infinity. To address this trade-off, we propose two novel estimators: the multi-kernel correntropy Kalman filter-based disturbance observer (MKCKF-DOB) and the interacting multiple models Kalman filter-based disturbance observer (IMMKF-DOB). Simulations verify the effectiveness of the proposed methods.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the bias - variance trade - off problem that exists when the perturbation model is incomplete during state and perturbation estimation using the Kalman - filter - based disturbance observer (KF - DOB). Specifically, the paper focuses on how to balance the bias and variance of the estimate to improve the estimation performance in the case where the perturbation model is partially known and there is random measurement noise.
### Background and Problem Description
In control and estimation systems, perturbations are a common problem, which can degrade the performance of the system or even lead to instability. Therefore, accurate perturbation estimation is crucial for many applications. Existing disturbance observers (DOB) can be roughly divided into frequency - domain methods and time - domain methods. Among them, the Kalman - filter - based disturbance observer (KF - DOB) can achieve the minimum mean - square - error estimate when the perturbation model is fully specified, but in practical applications, the perturbation model is often only partially known. This leads to the bias - variance trade - off problem in the KF - DOB estimation process.
### Main Contributions
1. **Reveal the Bias - Variance Trade - off**:
- The paper proves that in the presence of an inaccurate perturbation model and random noise, there is an inherent bias - variance trade - off in the KF - DOB.
- When the perturbation noise covariance approaches infinity, the KF - DOB is equivalent to simultaneous input and state estimation (SISE), thus highlighting the crucial role of the perturbation noise covariance in balancing the bias - variance effect.
2. **Propose Solutions**:
- The paper proposes two new estimators: the multi - kernel correntropy Kalman - filter disturbance observer (MKCKF - DOB) and the interacting multiple - model Kalman - filter disturbance observer (IMMKF - DOB) to alleviate the bias - variance trade - off problem.
### Methods and Results
1. **Bias - Variance Trade - off Analysis**:
- By analyzing the convergence rate and the steady - state error covariance when the initial value is incorrect, the paper explores in detail the influence of process covariance selection on the performance of the KF - DOB.
- The results show that a larger process covariance will lead to a faster bias convergence rate, but at the cost of an increased steady - state error covariance.
2. **Proposal of New Estimators**:
- **MKCKF - DOB**: Using the multi - kernel correntropy loss function, this estimator is robust to heavy - tailed processes or measurement noise.
- **IMMKF - DOB**: By dealing with Markov jump systems, this estimator can provide more accurate estimates in complex perturbation scenarios.
3. **Simulation Verification**:
- Simulation experiments verify the performance of the two proposed new estimators under different perturbation models, and the results show that they are superior to existing estimators in dealing with inaccurate perturbation models.
### Conclusion
Through theoretical analysis and experimental verification, the paper reveals the bias - variance trade - off problem of the KF - DOB under inaccurate perturbation models and proposes effective solutions. These results are helpful for better understanding the challenges of state estimation under inaccurate perturbation models and provide guidance for practical applications.