Convex Parameterizations and Fidelity Bounds for Nonlinear\n Identification and Reduced-Order Modelling

2017/01/23 by Mark M. Tobenkin, Ian R. Manchester, Tobenkin, Mark M. +3 · 1 citation
Physics and Astronomy · Engineering · #Model Reduction and Neural Networks #Control Systems and Identification #Advanced Control Systems Optimization

paper · pdf · doi:10.48550/arxiv.1701.06652

Abstract

Model instability and poor prediction of long-term behavior are common\nproblems when modeling dynamical systems using nonlinear "black-box"\ntechniques. Direct optimization of the long-term predictions, often called\nsimulation error minimization, leads to optimization problems that are\ngenerally non-convex in the model parameters and suffer from multiple local\nminima. In this work we present methods which address these problems through\nconvex optimization, based on Lagrangian relaxation, dissipation inequalities,\ncontraction theory, and semidefinite programming. We demonstrate the proposed\nmethods with a model order reduction task for electronic circuit design and the\nidentification of a pneumatic actuator from experiment.\n

Cited by

• State dimension reduction of recurrent equilibrium networks with contraction and robustness preservation

Related

• Stable Nonlinear Identification From Noisy Repeated Experiments via\n Convex Optimization
• Identification of Nonlinear Systems with Stable Limit Cycles via Convex Optimization
• Data-Driven Scenario Optimization for Automated Controller Tuning with\n Probabilistic Performance Guarantees
• Progressive construction of a parametric reduced-order model for PDE-constrained optimization
• Bounding Optimality Gaps for Non-Convex Optimization Problems: Applications to Nonlinear Safety-Critical Systems