Formulas for Data-driven Control: Stabilization, Optimality and Robustness

Claudio De Persis,Pietro Tesi

DOI: https://doi.org/10.48550/arXiv.1903.06842

2019-09-08

Abstract:In a paper by Willems and coauthors it was shown that persistently exciting data can be used to represent the input-output behavior of a linear system. Based on this fundamental result, we derive a parametrization of linear feedback systems that paves the way to solve important control problems using data-dependent Linear Matrix Inequalities only. The result is remarkable in that no explicit system's matrices identification is required. The examples of control problems we solve include the state and output feedback stabilization, and the linear quadratic regulation problem. We also discuss robustness to noise-corrupted measurements and show how the approach can be used to stabilize unstable equilibria of nonlinear systems.

Systems and Control,Optimization and Control

What problem does this paper attempt to address?

The paper attempts to address the problem of how to design controllers for control systems directly using data, without first identifying the parameters of the system model. Specifically, the paper focuses on how to solve the stability and optimal control problems of linear systems through data-driven methods and discusses the robustness of these methods in the presence of noise-corrupted measurements and nonlinear systems.

### Main Issues
1. **Stability of State Feedback and Output Feedback**: How to design state feedback and output feedback controllers using data-driven methods to ensure system stability.
2. **Linear Quadratic Regulation (LQR) Problem**: How to solve the LQR problem through data-driven methods, i.e., finding a state feedback controller that minimizes the performance index of the closed-loop system.
3. **Robustness to Noise-Corrupted Measurements**: How to design stable controllers when measurement data is corrupted by noise.
4. **Stability of Nonlinear Systems**: How to stabilize the unstable equilibrium points of nonlinear systems using data-driven methods.

### Solutions
The paper addresses the above issues based on the fundamental lemma proposed by Willems et al. through the following steps:

1. **Data-Dependent System Representation**:
- Using data-driven methods, represent the open-loop and closed-loop dynamics of the system through the Hankel matrix.
- Obtain estimates of the system matrices through the Least Squares Problem.

2. **Controller Parameterization**:
- Parameterize the controller as a function of the data using data-driven methods.
- Solve for the controller parameters through Linear Matrix Inequality (LMI) to ensure system stability.

3. **Robustness Analysis**:
- Discuss the robustness of data-driven methods in the presence of noise-corrupted measurements and nonlinear systems.
- Propose how to adjust the design conditions of LMI to ensure controller stability in the presence of noise.

### Main Contributions
1. **No Need for System Model Identification**: The proposed method does not require explicit identification of the system's parameter model and can design controllers directly from data.
2. **Stability Guarantee with Finite Data Points**: The stability of the controller can be guaranteed with a finite number of data points.
3. **Wide Application Prospects**: The method is not only applicable to linear systems but can also be extended to more complex design problems such as H∞ control and quadratic stabilization.

### Experimental Validation
The paper validates the effectiveness of the proposed method through numerical examples, including the control design of a batch reactor system, demonstrating the feasibility and superiority of the method in practical applications.

In summary, the paper provides a new approach to designing control systems through data-driven methods, particularly suitable for situations where system model identification is difficult or time-consuming.