Luke Bhan,Yuanyuan Shi,Iasson Karafyllis,Miroslav Krstic,James B. Rawlings
Abstract:Observers for PDEs are themselves PDEs. Therefore, producing real time estimates with such observers is computationally burdensome. For both finite-dimensional and ODE systems, moving-horizon estimators (MHE) are operators whose output is the state estimate, while their inputs are the initial state estimate at the beginning of the horizon as well as the measured output and input signals over the moving time horizon. In this paper we introduce MHEs for PDEs which remove the need for a numerical solution of an observer PDE in real time. We accomplish this using the PDE backstepping method which, for certain classes of both hyperbolic and parabolic PDEs, produces moving-horizon state estimates explicitly. Precisely, to explicitly produce the state estimates, we employ a backstepping transformation of a hard-to-solve observer PDE into a target observer PDE, which is explicitly solvable. The MHEs we propose are not new observer designs but simply the explicit MHE realizations, over a moving horizon of arbitrary length, of the existing backstepping observers. Our PDE MHEs lack the optimality of the MHEs that arose as duals of MPC, but they are given explicitly, even for PDEs. In the paper we provide explicit formulae for MHEs for both hyperbolic and parabolic PDEs, as well as simulation results that illustrate theoretically guaranteed convergence of the MHEs.
Systems and Control,Artificial Intelligence,Analysis of PDEs,Dynamical Systems,Optimization and Control
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is: **How to design a moving - horizon estimator (MHE) for partial differential equation (PDE) systems to avoid the high computational burden when solving PDE observers in real - time applications**.
Specifically, the paper introduces explicit MHE methods for two types of one - dimensional PDE systems (hyperbolic and parabolic). By using the PDE backstepping method, the difficult - to - solve observer PDE is transformed into an easy - to - solve target observer PDE, so that state estimates can be explicitly generated within a moving time window without complex numerical solutions. This method not only improves computational efficiency but also ensures the exponential convergence of the estimation error.
### The main contributions of the paper include:
1. **Proposing a new MHE method**: For hyperbolic and parabolic PDE systems, the paper introduces explicit MHE methods that can directly generate state estimates within a moving time window without the need to solve PDE observers in real - time.
2. **Ensuring the exponential convergence of the estimation error**: Through Lyapunov analysis, it is proved that the proposed MHE method can ensure that the estimation error converges exponentially to the true system state.
3. **Providing specific formulas and simulation results**: The paper gives specific formulas for MHE for hyperbolic and parabolic PDE systems and verifies the validity of the theoretical results through simulation experiments.
### Key technical means:
- **Backstepping Method**: Transform complex observer PDEs into easy - to - solve target PDEs through backstepping transformation.
- **Moving - Horizon Estimator (MHE)**: Use measurement data within a moving time window to generate state estimates, avoiding dependence on the entire historical data.
### Formula examples:
- For hyperbolic PDE systems, the error system of the observer can be expressed as:
\[
\tilde{u}_t(x, t) = \tilde{u}_x(x, t) + g(x)\tilde{u}(0, t) + \int_0^x f(x, y)\tilde{u}(y, t) dy - p_1(x)\tilde{u}(0, t)
\]
where \(\tilde{u}(x, t) = u(x, t) - \hat{u}(x, t)\) is the state estimation error and \(p_1(x)\) is the observer gain function.
- For parabolic PDE systems, the explicit form of MHE is:
\[
\hat{u}(x, t) = \hat{w}(x, t) - \int_x^1 k(y, x) \hat{w}(y, t) dy
\]
where the expression of \(\hat{w}(x, t)\) involves infinite series and integrals and has a relatively complex specific form.
Through these methods, the paper successfully solves the computational burden problem of PDE observers in real - time applications and provides an efficient and accurate method for state estimation of PDE systems.