Abstract:This paper studies the problem of online stabilization of an unknown discrete-time linear time-varying (LTV) system under bounded non-stochastic (potentially adversarial) disturbances. We propose a novel control algorithm based on convex body chasing (CBC). Under the assumption of infrequently changing or slowly drifting dynamics, the algorithm guarantees bounded-input-bounded-output stability in the closed loop. Our approach avoids system identification and applies, with minimal disturbance assumptions, to a variety of LTV systems of practical importance. We demonstrate the algorithm numerically on examples of LTV systems including Markov linear jump systems with finitely many jumps.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to perform online stable control of an unknown linear time - varying system (LTV) in the presence of bounded and possibly adversarial disturbances. Specifically, the researchers focus on how to design an algorithm to ensure the bounded - input - bounded - output (BIBO) stability of the system without fully understanding the dynamic parameters of the system. This involves several key challenges:
1. **Unknown system dynamics**: Unlike traditional control methods based on known models, this paper deals with the situation where the system dynamic parameters are unknown. This means that the controller must be able to adapt to the changes of the system over time without prior knowledge of the specific form or range of these changes.
2. **Adversarial disturbances**: In addition to the uncertainty of the system dynamics, the system is also affected by bounded but possibly adversarial disturbances. Such disturbances may come from the external environment or internal factors, increasing the difficulty of control.
3. **Online learning and control**: The method proposed by the researchers needs to be able to learn and adjust the control strategy in real - time during the operation of the system to cope with the effects of dynamic changes and disturbances. This method requires that the algorithm not only has good learning ability but also has the ability to respond quickly to maintain the stability of the system.
To address the above challenges, the paper proposes a new algorithm based on the Convex Body Chasing (CBC) technique. CBC is an online learning method for selecting a sequence of points such that the path length of these points in continuously presented convex sets is minimized. Through this method, the algorithm can select a hypothesized model consistent with the observed data without knowing the exact model of the system, and synthesize a deterministic - equivalent linear quadratic regulator (LQR) controller based on this hypothesized model, thereby achieving stable control of the system.
### Main contributions
- **BIBO stability guarantee**: The paper proves that in the case of limited or sub - linear growth of system dynamic changes, the proposed algorithm can ensure the BIBO stability of the system. This is achieved by analyzing the competitive ratio of the CBC technique.
- **No need for system identification**: Compared with traditional methods, this algorithm does not require explicit identification of the system, thus avoiding stability problems caused by system identification errors.
- **Applicable to a wide range of practical scenarios**: The paper demonstrates the effectiveness of the algorithm in various LTV systems through numerical experiments, including Markov linear jump systems and systems with slowly drifting dynamics.
### Numerical experiments
The paper verifies the effectiveness of the algorithm through two specific LTV system examples:
1. **Markov linear jump system**: In this example, the system randomly switches between two different states and is subject to bounded adversarial disturbances. The experimental results show that compared with the method based on online least squares (OLS), the proposed algorithm can successfully stabilize the system under a wider range of parameter settings and has a smaller state norm.
2. **LTV system**: In this example, the system has slowly drifting dynamic characteristics. The experimental results show that the performance of the proposed algorithm in the data collection phase is better than that of the random noise injection method, especially in the case of an unstable system.
In conclusion, the paper proposes a novel online control algorithm that can achieve stable control in an unknown linear time - varying system and has important theoretical and practical application values.