Rapid Accelerator Simulations Using PARFOR

This example shows the use of Rapid Accelerator in applications that require running parallel simulations for a range of input and parameter values.

We use the engine idle speed model which simulates the idle speed of an engine. The input of this model is the voltage of the bypass air valve and the output is the idle speed.

We run parallel simulations using PARFOR with two sets of valve voltages and by independently varying two of the three gain parameters of the transfer functions over a range of two values. Hence, in total, we will be running eight different sets of simulations.

It is easy to customize this example for your own application by modifying the script file used to build this example. Click the link in the top left corner of this page to edit the script file. Click the link in the top right corner to run this example from MATLAB®. Before running this example, make sure you are in a writable directory.

Step 1: Preparation

First we open the model where the simulation mode has been set to Rapid Accelerator. The default input data and the required parameters are preloaded in the models workspace.

The parameters gain2 and gain3 have been specified as tunable parameters so that they can be modified later using the utility function Simulink.BlockDiagram.modifyTunableParameters. To learn how to select tunable parameters and set their properties graphically, read the help page concerning the Model Parameter Configuration Dialog Box.

We copy the default input and time data to a variable so that we can later modify them and pass them to the SIM command.

% Open model:
mdl = 'sldemo_raccel_engine_idle_speed';

% Copy input data
inpData = evalin('base', 'inpData');
tData = evalin('base', 'time');

Step 2: Build the Rapid Accelerator Target

First, make sure you are in a writable directory because in this step we will generate extra files. We then build the Rapid Accelerator executable for the model and get the default run-time parameter set.

rtp = Simulink.BlockDiagram.buildRapidAcceleratorTarget(mdl);
close_system(mdl, 0);
### Building the rapid accelerator target for model: sldemo_raccel_engine_idle_speed
### Successfully built the rapid accelerator target for model: sldemo_raccel_engine_idle_speed

Step 3: Create Parameter Sets

Using the default rtp structure from step 2, we build a new structure with different values for the tunable variables in the model. We want to see how the idle speed changes for different values of parameters gain2 and gain3. Therefore, we generate different parameter sets with different values of gain2 and gain3 and leave the other tunable variables at their default values.

The utility function Simulink.BlockDiagram.modifyTunableParameters is a convenient way to build the rtp structure with different parameter values.

gain2_vals = 25:10:35;
gain3_vals = 20:10:30;

num_gain2_vals = length(gain2_vals);
num_gain3_vals = length(gain3_vals);
numParamSets = num_gain2_vals*num_gain3_vals;

% Create parameter sets:
paramSets = cell(1, numParamSets);
idx = 1;
for iG2 = 1:num_gain2_vals
    for iG3 = 1:num_gain3_vals
        paramSets{idx} = ...
            Simulink.BlockDiagram.modifyTunableParameters(rtp, ...
            'gain2',gain2_vals(iG2), ...
        idx = idx+1;

Step 4: Create Input Sets

Here we perturb the default input values vector to obtain a new input values vector.

In this example, we will be plotting the engine idle speed as a function of the valve voltage for different parameter values.

inpSets{1} = inpData;
rndPertb = 0.5 + rand(length(tData), 1);
inpSets{2} = inpSets{1}.*rndPertb;
numInpSets  = length(inpSets);

Step 5: Create SIM Command Argument Sets

We now create a cell array of parameter-name-value structures that will be passed to the SIM command called from inside of a PARFOR loop.

To run the SIM command in Rapid Accelerator mode, we need to set the field 'RapidAcceleratorUpToDateCheck' to 'off' and pass the parameter sets by using the 'RapidAcceleratorParameterSets' field.

We also collect all of the external inputs in a cell array. Later we assign each of them with 'externalInput' as the variable name in the base workspace of the workers.

numSimCmdArgStructs = numParamSets*numInpSets;
simCmdParamValStructs = cell(1, numSimCmdArgStructs);
externalInput = cell(1, numSimCmdArgStructs);

paramValStruct.SaveTime = 'on';
paramValStruct.SaveOutput = 'on';
paramValStruct.LoadExternalInput = 'on';
% 'externalInput' is the name of the base workspace variable of
% the MATLAB worker sessions containing the external inputs data
paramValStruct.ExternalInput = 'externalInput';
paramValStruct.SimulationMode = 'rapid';
paramValStruct.RapidAcceleratorUpToDateCheck = 'off';
paramValStruct.RapidAcceleratorParameterSets = [];

idx = 1;
for paramSetsIdx = 1:numParamSets
    for inpSetsIdx = 1:numInpSets
        simCmdParamValStructs{idx} = paramValStruct;
        simCmdParamValStructs{idx}.RapidAcceleratorParameterSets = ...
        externalInput{idx} = [tData, inpSets{inpSetsIdx}];
        idx = idx + 1;

Step 6: Start parpool

Uncomment the code to start a parpool The following line of code starts four worker MATLAB sessions. PARFOR would then distribute jobs to these four worker sessions.

% parpool('local');

Step 7: Simulate in PARFOR

Here we simulate the model in parallel using PARFOR with different argument sets that contain different parameter values and input vectors. We assign the input vectors corresponding to the simulation in the base workspace of the MATLAB worker session that is running the simulation. The use of EVALIN('base') and ASSIGNIN('base') inside of a PARFOR loop indicates a reference to the base workspaces of the worker machines, and thus is not generally recommended. However, in this example, the variable 'externalInputs' is required by the base workspace of each session. Consequently, using ASSIGNIN('base') inside of PARFOR is valid here.

out = cell(1, numSimCmdArgStructs);

parfor(i = 1:numSimCmdArgStructs)
    assignin('base', 'externalInput', externalInput{i}); %#ok<PFEVB>
    out{i} = sim(mdl, simCmdParamValStructs{i});
Starting parallel pool (parpool) using the 'local' profile ... connected to 12 workers.

Step 8: Plot Results

We now plot the engine idle speed with respect to time for different parameter values and inputs. The variable 'out' is a cell array of Simulink.SimulationOutput objects which contains the simulation data for each simulation.

for i=1:numSimCmdArgStructs
    t = out{i}.find('tout');
    y = out{i}.find('yout');
    plot(t, y)
    hold all

fprintf('\n Contents of the out{1}: \n');
 Contents of the out{1}: 


            tout: [1041×1 double]
       ScopeData: [1041×2 double]
    valveVoltage: [1041×1 double]
            yout: [1041×1 double]

Use 'get' to access a variable by name.
Use 'getSimulationMetadata' to access metadata about the simulation.

Step 9: Close parpool

If the parpool was started earlier then it must be closed. The second line of the following code closes the parpool, and thus closes the worker sessions, when the comment symbol is removed. For more information, see the documentation on PARFOR and PARPOOL.

% delete(gcp('nocreate'))
Was this topic helpful?