This is machine translation

Translated by Microsoft
Mouseover text to see original. Click the button below to return to the English version of the page.

Note: This page has been translated by MathWorks. Click here to see
To view all translated materials including this page, select Country from the country navigator on the bottom of this page.

Generate Code To Compute Optimal MPC Moves in MATLAB

This example shows how to use the mpcmoveCodeGeneration command to generate C code to compute optimal MPC control moves for real-time applications.

Plant Model

The plant is a single-input, single-output, stable, 2nd order linear plant.

plant = tf(5,[1 0.8 3]);

Convert the plant to discrete-time, state-space form, and specify a zero initial states vector.

Ts = 1;      
plant = ss(c2d(plant,Ts));
x0 = zeros(size(plant.B,1),1);

Design MPC Controller

Create an MPC controller with default horizons.

mpcobj = mpc(plant,Ts);
-->The "PredictionHorizon" property of "mpc" object is empty. Trying PredictionHorizon = 10.
-->The "ControlHorizon" property of the "mpc" object is empty. Assuming 2.
-->The "Weights.ManipulatedVariables" property of "mpc" object is empty. Assuming default 0.00000.
-->The "Weights.ManipulatedVariablesRate" property of "mpc" object is empty. Assuming default 0.10000.
-->The "Weights.OutputVariables" property of "mpc" object is empty. Assuming default 1.00000.

Specify controller tuning weights.

mpcobj.Weights.MV = 0;
mpcobj.Weights.MVrate = 0.5;
mpcobj.Weights.OV = 1;

Specify initial constraints on the manipulated variable and plant output. These constraints will be updated at run time.

mpcobj.MV.Min = -1;
mpcobj.MV.Max = 1;
mpcobj.OV.Min = -1;
mpcobj.OV.Max = 1;

Simulate Online Constraint Changes with mpcmove Command

In the closed-loop simulation, constraints are updated and fed into the mpcmove command at each control interval.

yMPCMOVE = [];
uMPCMOVE = [];

Set the simulation time.

Tsim = 20;

Initialize the online constraint data.

MVMinData = -0.2-[1 0.95 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 ...
    0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1];
MVMaxData = 0.2+[1 0.95 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 ...
    0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1];
OVMinData = -0.2-[1 0.95 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 ...
    0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1];
OVMaxData = 0.2+[1 0.95 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 ...
    0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1];

Initialize plant states.

x = x0;

Initialize MPC states.

xmpc = mpcstate(mpcobj);
-->Assuming output disturbance added to measured output channel #1 is integrated white noise.
-->The "Model.Noise" property of the "mpc" object is empty. Assuming white noise on each measured output channel.

Run a closed-loop simulation by calling mpcmove in a loop.

options = mpcmoveopt;
for ct = 1:round(Tsim/Ts)+1
    % Update and store plant output.
    y = plant.C*x;
    yMPCMOVE = [yMPCMOVE y];
    % Update constraints.
    options.MVMin = MVMinData(ct);
    options.MVMax = MVMaxData(ct);
    options.OutputMin = OVMinData(ct);
    options.OutputMax = OVMaxData(ct);
    % Compute control actions.
    u = mpcmove(mpcobj,xmpc,y,1,[],options);
    % Update and store plant state.
    x = plant.A*x + plant.B*u;
    uMPCMOVE = [uMPCMOVE u];
end

Validate Simulation Results with mpcmoveCodeGeneration Command

To prepare for generating code that computes optimal control moves from MATLAB, it is recommended to reproduce the same control results with the mpcmoveCodeGeneration command before using the codegen command from the MATLAB Coder product.

yCodeGen = [];
uCodeGen = [];

Initialize plant states.

x = x0;

Create data structures to use with mpcmoveCodeGeneration using getCodeGenerationData.

[coredata,statedata,onlinedata] = getCodeGenerationData(mpcobj);

Run a closed-loop simulation by calling mpcmoveCodeGeneration in a loop.

for ct = 1:round(Tsim/Ts)+1
    % Update and store plant output.
    y = plant.C*x;
    yCodeGen = [yCodeGen y];
    % Update measured output in online data.
    onlinedata.signals.ym = y;    
    % Update reference in online data.
    onlinedata.signals.ref = 1;    
    % Update constraints in online data.
    onlinedata.limits.umin = MVMinData(ct);
    onlinedata.limits.umax = MVMaxData(ct);
    onlinedata.limits.ymin = OVMinData(ct);
    onlinedata.limits.ymax = OVMaxData(ct);
    % Compute control actions.
    [u,statedata] = mpcmoveCodeGeneration(coredata,statedata,onlinedata);
    % Update and store plant state.
    x = plant.A*x + plant.B*u;
    uCodeGen = [uCodeGen u];
end

The simulation results are identical to those using mpcmove.

t = 0:Ts:Tsim;
figure;
subplot(1,2,1)
plot(t,yMPCMOVE,'--*',t,yCodeGen,'o');
grid
legend('mpcmove','codegen')
title('Plant Output')
subplot(1,2,2)
plot(t,uMPCMOVE,'--*',t,uCodeGen,'o');
grid
legend('mpcmove','codegen')
title('Controller Moves')

Generate MEX Function From mpcmoveCodeGeneration Command

To generate C code from the mpcmoveCodeGeneration command, use the codegen command from the MATLAB Coder product. In this example, generate a MEX function mpcmoveMEX to reproduce the simulation results in MATLAB. You can change the code generation target to C/C++ static library, dynamic library, executable, etc. by using a different set of coder.config settings.

When generating C code for the mpcmoveCodeGeneration command:

  • Since no data integrity checks are performed on the input arguments, you must make sure that all the input data has the correct types, dimensions, and values.

  • You must define the first input argument, mpcmove_struct, as a constant when using the codegen command.

  • The second input argument, mpcmove_state, is updated by the command and returned as the second output. In most cases, you do not need to modify its contents and should simply pass it back to the command in the next control interval. The only exception is when custom state estimation is enabled, in which case you must provide the current state estimation using this argument.

if ~license ('test', 'MATLAB_Coder')
    disp('MATLAB Coder(TM) is required to run this example.')
    return
end

Generate MEX function.

fun = 'mpcmoveCodeGeneration';
funOutput = 'mpcmoveMEX';
Cfg = coder.config('mex');
Cfg.DynamicMemoryAllocation = 'off';
codegen('-config',Cfg,fun,'-o',funOutput,'-args',...
    {coder.Constant(coredata),statedata,onlinedata});

Initialize data storage.

yMEX = [];
uMEX = [];

Initialize plant states.

x = x0;

Use getCodeGenerationData to create data structures to use with mpcmoveCodeGeneration.

[coredata,statedata,onlinedata] = getCodeGenerationData(mpcobj);

Run a closed-loop simulation by calling the generated mpcmoveMEX functions in a loop.

for ct = 1:round(Tsim/Ts)+1
    % Update and store the plant output.
    y = plant.C*x;
    yMEX = [yMEX y];
    % Update measured output in online data.
    onlinedata.signals.ym = y;    
    % Update reference in online data.
    onlinedata.signals.ref = 1;    
    % Update constraints in online data.
    onlinedata.limits.umin = MVMinData(ct);
    onlinedata.limits.umax = MVMaxData(ct);
    onlinedata.limits.ymin = OVMinData(ct);
    onlinedata.limits.ymax = OVMaxData(ct);
    % Compute control actions.
    [u,statedata] = mpcmoveMEX(coredata,statedata,onlinedata);
    % Update and store the plant state.
    x = plant.A*x + plant.B*u;
    uMEX = [uMEX u];
end

The simulation results are identical to those using mpcmove.

figure
subplot(1,2,1)
plot(t,yMPCMOVE,'--*',t,yMEX,'o')
grid
legend('mpcmove','mex')
title('Plant Output')
subplot(1,2,2)
plot(t,uMPCMOVE,'--*',t,uMEX,'o')
grid
legend('mpcmove','mex')
title('Controller Moves')

See Also

|

Related Topics