Ioannis A. Fotiou,Philipp Rostalski,Bernd Sturmfels,Manfred Morari
Abstract:We present a method for nonlinear parametric optimization based on algebraic geometry. The problem to be studied, which arises in optimal control, is to minimize a polynomial function with parameters subject to semialgebraic constraints. The method uses Groebner bases computation in conjunction with the eigenvalue method for solving systems of polynomial equations. In this way, certain companion matrices are constructed off-line. Then, given the parameter value, an on-line algorithm is used to efficiently obtain the optimizer of the original optimization problem in real time.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to effectively solve nonlinear polynomial optimization problems with parameters in optimal control. Specifically, the paper focuses on how to quickly find the optimal solution under given parameter values, so as to achieve real - time control.
### Problem Background
In the field of optimal control, especially in model predictive control (MPC), it is usually necessary to solve an optimization problem online to obtain the optimal control input. For nonlinear systems, this optimization problem is often a nonlinear polynomial optimization problem with parameters. Traditional online solution methods have a large amount of computation and are difficult to meet the requirements of real - time performance.
### Core Contributions of the Paper
To solve this problem, the paper proposes a method based on algebraic geometry, using Gröbner bases and eigenvalue methods to solve nonlinear polynomial optimization problems with parameters. The specific steps are as follows:
1. **Offline Computation**:
- For a given optimization problem, construct an ideal under KKT conditions and calculate its Gröbner basis.
- Use the Gröbner basis to construct generalized companion matrices. The eigenvalues of these matrices correspond to the solutions of the optimization problem.
- Through the analysis of these companion matrices, pre - calculate all possible candidate solutions.
2. **Online Computation**:
- Given specific parameter values, quickly find the optimal solution by solving the eigenvalue problem of the companion matrices.
- Check the feasibility of the solution and select the feasible solution that minimizes the objective function.
### Formula Summary
- **Form of the Optimization Problem**:
\[
\min_{u} J(u, x) \quad \text{s.t.} \quad g(u, x) \leq 0
\]
where \( J(u, x) \) is the objective function and \( g(u, x) \) is the constraint condition.
- **KKT Conditions**:
\[
\nabla_u J(u, x) + \sum_{i = 1}^q \mu_i \nabla_u g_i(u, x)=0
\]
\[
\mu_i g_i(u, x) = 0, \quad \mu_i\geq0, \quad g(u, x)\leq0
\]
- **Generalized Companion Matrix**:
\[
M_h=[a_{ij}]=\begin{bmatrix}
a_1^T \\
\vdots \\
a_l^T
\end{bmatrix}
\]
where \( a_i \) is the coefficient vector of the standard monomial.
### Application Example
The paper shows the application of this method through the example of the Duffing oscillator. By pre - processing the generalized companion matrices calculated offline, the optimal control input can be quickly solved in the online stage, so as to achieve stable control of the system.
### Conclusion
The method based on algebraic geometry proposed in this paper significantly improves the solution efficiency of nonlinear polynomial optimization problems with parameters through offline pre - calculation and online fast solution, and is especially suitable for real - time optimal control applications.