POD-Galerkin model order reduction for parametrized nonlinear time-dependent optimal flow control: an application to shallow water equations

Maria Strazzullo, Francesco Ballarin, Gianluigi Rozza

TL;DR

This paper introduces a novel approach using Proper Orthogonal Decomposition (POD) and Galerkin projection to efficiently solve complex, time-dependent optimal control problems governed by shallow water equations. This method significantly speeds up simulations for environmental monitoring and coastal engineering applications by creating accurate reduced-order models.

What Problem Does This Paper Address?

Simulating parametrized optimal control problems (OCPs) governed by partial differential equations (PDEs) is crucial for many scientific fields, especially in environmental sciences for applications like marine ecosystem management. However, these simulations, particularly those involving nonlinear, time-dependent models like the shallow water equations (SWEs), are computationally very expensive. This high cost limits the ability to explore a wide range of parameter configurations or perform real-time analysis. The authors aim to bridge this gap by developing a method that can provide fast yet accurate solutions for these complex OCPs.

Methods

The core of this work lies in applying Reduced Order Methods (ROMs), specifically a POD-Galerkin approach, to a nonlinear, time-dependent OCP constrained by SWEs. The methodology follows an "optimize-then-discretize" strategy.

  1. Optimality System Formulation: They first derive the optimality system for the SWEs-governed OCP using a Lagrangian approach. This system includes state, adjoint, and optimality equations.
  2. Space-Time Discretization: A full-order approximation of this optimality system is obtained using a space-time Finite Element (FE) discretization. This results in a large, nonlinear system of equations.
  3. POD-Galerkin Reduction: To tackle the computational burden, they employ a partitioned POD-Galerkin approach. This involves:
  4. Snapshot Generation: Computing truth solutions for a set of parameters from the parameter space.
  5. POD Basis Construction: Applying POD to these snapshots to generate reduced-order basis functions for each variable (velocity, elevation, control, adjoint velocity, adjoint elevation).
  6. Galerkin Projection: Projecting the full-order optimality system onto these reduced spaces, resulting in a significantly smaller, computationally tractable system.
  7. Aggregated Spaces: To ensure the solvability of the reduced saddle-point system, they utilize an "aggregated spaces" technique, which builds a common space for state and adjoint variables.
  8. Offline-Online Decomposition: The method is structured for an efficient offline-online decomposition. The offline stage involves assembling parameter-independent quantities and building reduced spaces, while the online stage efficiently solves the reduced system for new parameter values.

Datasets

The paper uses a synthetic test case inspired by existing literature ([25]) for the shallow water equations. This test case involves simulating the spreading of a Gaussian-distributed water mass with an initial null velocity. The parameters explored are related to diffusion, advection, and desired state profiles. No new datasets were introduced; rather, a specific configuration of the SWEs was used for validation.

Key Insights and Ideas

  • ROMs for Complex OCPs: The paper demonstrates that ROMs, particularly POD-Galerkin, are effective even for highly complex, nonlinear, time-dependent OCPs, which were previously considered too challenging for such methods.
  • Partitioned POD: Applying POD separately to each variable (state velocity, state elevation, control, adjoint velocity, adjoint elevation) is a viable strategy for this problem.
  • Aggregated Spaces for Solvability: The use of aggregated spaces is crucial for maintaining the saddle-point structure and guaranteeing the solvability of the reduced optimality system, a common challenge in OCPs.
  • Speedup Potential: The methodology offers a significant speedup, enabling many simulations in the time it takes to perform one full-order simulation, which is vital for parameter exploration and real-time applications.

Key Results

  • Accuracy: The ROM approximation achieved high accuracy, with average relative errors reaching $10^{-3}$ for state and adjoint velocity and $10^{-4}$ for state and adjoint elevation, and control, when using a reduced basis of size $N=30$.
  • Speedup: The paper reports a speedup index of approximately 30, meaning that up to 30 reduced simulations can be performed in the time of one full-order simulation. This speedup is relatively insensitive to the basis size $N$.
  • Visual Validation: Figures 3-6 show a strong qualitative agreement between the full-order (FE) and reduced-order (ROM) solutions for state and adjoint velocity and elevation profiles at various time steps, demonstrating the method's ability to capture the dynamics.

Limitations and Future Work

  • Theoretical Analysis: The authors acknowledge that a more rigorous theoretical analysis of the considered problem, especially concerning the hyperbolic nature of SWEs and potential ROM approximation issues, is still needed.
  • Error Estimator: The current approach relies on a heuristic choice of snapshots for POD. The development of an error estimator would allow for more efficient greedy-based ROM algorithms.
  • Random Parameters: For environmental applications, extending the framework to handle random input parameters (beyond deterministic ones) is suggested as a future direction.
  • Computational Cost: While significantly reduced, the offline stage of ROM construction can still be computationally intensive, especially with a large number of snapshots. The paper mentions that the choice of $N_{max}$ (number of snapshots) affects this effort.

Summary updated

Abstract

In this work we propose reduced order methods as a reliable strategy to efficiently solve parametrized optimal control problems governed by shallow waters equations in a solution tracking setting. The physical parametrized model we deal with is nonlinear and time dependent: this leads to very time consuming simulations which can be unbearable e.g. in a marine environmental monitoring plan application. Our aim is to show how reduced order modelling could help in studying different configurations and phenomena in a fast way. After building the optimality system, we rely on a POD-Galerkin reduction in order to solve the optimal control problem in a low dimensional reduced space. The presented theoretical framework is actually suited to general nonlinear time dependent optimal control problems. The proposed methodology is finally tested with a numerical experiment: the reduced optimal control problem governed by shallow waters equations reproduces the desired velocity and height profiles faster than the standard model, still remaining accurate.
BibTeX
@article{strazzullo2022pod,
  author = {Maria Strazzullo and Francesco Ballarin and Gianluigi Rozza},
  title = {{POD-Galerkin model order reduction for parametrized nonlinear time-dependent optimal flow control: an application to shallow water equations}},
  journal = {J. Num. Math},
  year = {2022},
  doi = {10.1515/jnma-2020-0098},
}
Figure