40.019 Dynamical Systems, Decisions, and Control

This course introduces the mathematical and computational foundations of dynamic decision-making, focusing on the modelling, analysis, control, and learning of dynamical systems. Students first develop fundamental concepts in state-space modelling, nonlinear dynamics, stability, controllability, observability, and feedback control. Building on these foundations, the course introduces stochastic dynamical systems, optimal control using Linear Quadratic Regulators (LQR), and state estimation using Kalman filters. The course then explores modern data-driven approaches for learning dynamical models from observations, including system identification, Dynamic Mode Decomposition (DMD), Sparse Identification of Nonlinear Dynamics (SINDy), and Gaussian Process Regression. Finally, students examine uncertainty quantification and predictive guarantees for learned dynamical models. Throughout the course, theoretical concepts are complemented by computational examples and applications to real-world engineering systems.

Learning objectives
  • Learn state-space model representations of dynamical systems, and core system properties (equilibria, stability, observability, and controllability).
  • Examine nonlinear state-space models to find equilibria (fixed-points), assess their stability, and linearize the system.
  • Apply model-based tests to assess stability, observability, and controllability of linear dynamical systems.
  • Design state observers and feedback controllers for linear systems using pole-placement techniques.
  • Analyse stochastic dynamical systems and design optimal feedback controllers using Linear Quadratic Regulator (LQR) methods.
  • Design Kalman filters for state estimation in stochastic dynamical systems and evaluate their estimation performance.
  • Apply data-driven system identification methods, including classical regression, Dynamic Mode Decomposition (DMD), Sparse Identification of Nonlinear Dynamics (SINDy), and Gaussian Process Regression (GPR), to learn dynamical models from input-output data.
Measurable outcomes
  • Explain the structure of state-space representations for discrete-time dynamical systems, including state variables, inputs, outputs, and transition functions and matrices.
  • Compute equilibrium points of nonlinear state-space models, determine their stability, and derive linearized approximations around equilibrium points.
  • Determine system properties such as stability, controllability, and observability of dynamical systems from linear state-space representations.
  • Design linear state-feedback controllers and Luenberger observers to modify system behaviour and estimate states from available outputs, respectively.
  • Implement and analyse state-space models, feedback controllers, and observers using MATLAB/Python and interpret the results.
  • Formulate stochastic state-space models and compute optimal feedback control laws using Linear Quadratic Regulator (LQR) methods, including the solution of algebraic Riccati equations.
  • Design steady-state Kalman filters for stochastic dynamical systems, compute estimation error covariances, and evaluate the performance of state estimators under process and measurement uncertainty.
  • Implement and compare data-driven system identification methods, including regression-based approaches, Dynamic Mode Decomposition (DMD), Sparse Identification of Nonlinear Dynamics (SINDy), and Gaussian Process Regression (GPR), to estimate dynamical models from input-output data.
Prerequisites
Course instructor

 

 

Number of credits: 12