LIBRARY

MPCComponents

Documentation for MPCComponents.

MPCComponents is a library of model-predictive control components implemented in the Dyad language. Linear MPC is backed by LinearMPC.jl, nonlinear MPC by acados, nonlinear state estimation by LowLevelParticleFilters.jl, and a purely discrete model built from these components can be exported as standalone C code.

The controllers and observers are clocked components: they are connected to a continuous-time plant through samplers and holds from DiscreteComponents, and a closed-loop model is simulated like any other Dyad model.

Component	Description
LinearMPCController	Linear MPC as one clocked component: state observer and condensed QP.
LinearMPCObserver, LinearMPCOptimizer	The two parts of LinearMPCController, also usable on their own.
PIDEquivalentMPCController	Reproduces DiscreteComponents.DiscretePIDStandard exactly while no constraint is active, so that constraints and preview can be added to a tuned PID controller.
ACADOSMPC	Nonlinear MPC for ODE, semi-explicit index-1 DAE and discrete-time prediction models.
UKFObserver	Unscented Kalman filter for a nonlinear prediction model, as one clocked component.
DelayLine	Tapped delay line, used to assemble preview vectors.

The analysis CCodeExport exports a model containing these components to C.

Installation

Install MPCComponents from the Dyad registry with

import Pkg
Pkg.add("MPCComponents")

acados is provided by acados_jll; neither a C compiler nor Python or CasADi is needed to simulate.

Getting started

• Model-predictive control: the receding-horizon principle, the signals of an MPC loop, a comparison with PID and LQR control, and the choice between linear and nonlinear MPC.

• Linear MPC: the prediction model, cost function, constraints, horizons and Kalman filter of LinearMPCController, with a reference-tracking example.

• Nonlinear MPC: the prediction model, discretization, cost function, constraints, solver and state estimation of ACADOSMPC, with a pendulum example.

Tutorials

The tutorials are translations of the MPC tutorials of DyadControlSystems.

• MPC with quadratic cost functions: nonlinear MPC with ACADOSMPC of a pendulum on a cart, with input bounds and an LQR terminal cost.

• Cart-pole MPC with a penalty on the control increments: regulation of the full state of a pendulum on a cart to the origin with ACADOSMPC, with a hard force bound and a soft angular-velocity bound.

• Disturbance modeling and rejection: rejection of a load disturbance with LinearMPCController on a first-order plant, without a disturbance model and with an input-disturbance model in the Kalman filter at two estimation speeds.

• MPC with binary or integer variables (MINLP): a double integrator whose control is restricted to  , solved as a mixed-integer QP by LinearMPCOptimizer, with a shortened integer horizon.

• MPC control of a Continuously Stirred Tank Reactor (CSTR): nonlinear MPC with ACADOSMPC of a reactor with hard input bounds and soft temperature bounds, and the effect of a plant-model mismatch.

• Autonomous lane changes: nonlinear MPC with ACADOSMPC of a kinematic bicycle model that changes lanes, with a smooth lateral reference, bounds on the state and the inputs, and an increment penalty.

• Economic MPC for residential HVAC: LinearMPCController minimizes the electricity cost of a house heated by a heat pump, using 24 h forecasts of the electricity price and the outdoor temperature, compared with a controller that only maintains comfort.

• Control design for a quadruple-tank system: linear MPC (LinearMPCController with a Kalman filter) and nonlinear MPC (ACADOSMPC) of a non-minimum-phase process with bounds on the pump voltages and the tank levels.

The component reference at the bottom of the sidebar lists every component, type, analysis and example of the library.