MATLAB: Model Predictive Control Toolbox

model predictive controlModel Predictive Control Toolboxsimulink

Dear all,

The problem is about "MPC block"(model predictive control) in MatLAB 7.0.

I drew the diagram of the control system in simulink. I added the "MPC block" from "Model Predictive Control Toolbox" to my control system.

After that, I double clicked the MPC block, one dialog box named "MPC Simulink Block" showed up.

Then I pressed the button "design".

After I pressed the button "design", MATLAB began to "build MPC Design Project".

After a while, a message box showed up and told me: "Could not assign linearization i/o. Message returned was:Error using==> mpcblockiotable Linearization mismatch between block i/o and linearization i/o"

I don't know why.

Best Answer

The question author provided me with the model to look at. The problem in this case happens due to the way plant model is implemented. It is done using level-1 MATLAB s-functions. The way these functions are written prevents them from linearizing correctly, causing the error message.

The recommendation is to use standard state-space and transfer function blocks for the plant model if possible.

Arkadiy

Related Solutions

MATLAB: Are the nominal conditions of the nonlinear CSTR model set to [0 0 u] in the example code for successive linearization in the Model Predictive Control Toolbox Version 2.2 (R14SP2)

This enhancement has been incorporated in Release 2008a (R2008a). For previous product releases, read below for any possible workarounds:

The reason why we have "Model.Nominal.U = [0 0 u];" instead of "Model.Nominal.U = up;" is that the first 2 plant inputs are designated as unmeasured disturbance to the Model Predictive Control (MPC) Toolbox by the following structure in the example code:

Plant.InputGroup.UnmeasuredDisturbances = [1 2];

This implies that the Model Predictive Control (MPC) toolbox doesn’t know the values of those 2 plant inputs during its optimization calculation. Therefore, by design, nominal values for all the unmeasured disturbance variables to the Model Predictive Control (MPC) toolbox have to be 0. In other words, if we have

Plant.InputGroup.UnmeasuredDisturbances = [1];
Plant.InputGroup.MeasuredDisturbances = [2];
Plant.InputGroup.ManipulatedVariables = [3];

Then we will see:

Model.Nominal.U = [0 up(2) up(3)];

Note that those two values are used by the plant itself in the closed loop simulation.

MATLAB: How to specify plant while using MPC Toolbox when I don’t have state space equations of the plant

To create a plant model, you can directly specify a linear model (state space equations), linearize a Simulink model, or identify a linear model using measured data. Following documentation explains this in detail:

https://www.mathworks.com/help/releases/R2016b/mpc/plant-specification.html

As described in the above documentation, you can use the MPC Designer App to linearize your plant and obtain the plant model:

https://www.mathworks.com/help/releases/R2016b/mpc/gs/linearize-simulink-models-using-mpc-designer.html

You can also use following method to linearize your Simulink model without using the MPC Designer:

https://www.mathworks.com/help/releases/R2016b/mpc/gs/using-simulink-to-develop-lti-models.html

If your model is strongly nonlinear and you wish to use another control mechanism you can explore Adaptive MPC or Gain-Scheduled MPC (Non-linear MPC is also available starting from R2017a):

https://www.mathworks.com/help/releases/R2016b/mpc/

Best Answer

Related Solutions

MATLAB: Are the nominal conditions of the nonlinear CSTR model set to [0 0 u] in the example code for successive linearization in the Model Predictive Control Toolbox Version 2.2 (R14SP2)

MATLAB: How to specify plant while using MPC Toolbox when I don’t have state space equations of the plant

Related Question