MATLAB Answers

0

Random Variable in Parallel Simulation

Asked by Baris Ciftci on 19 May 2018
Latest activity Answered by Rahul Kumar on 29 Aug 2018
Hi,
In my Simulink model, I have an Embedded MATLAB Function which generates random variables by rand command periodically during the simulation interval. Now, I want to run multiple simulations with parsim command. However, the same random variables are generated in different (multiple) runs of the model. I tried to add rng('shuffle') command into Embedded MATLAB Function, but it gave error saying it is not supported to use 'shuffle' command in Embedded MATLAB Function. Then, I tried to add rng(seed) command in Embedded MATLAB Function. rng(seed) works fine in Embedded MATLAB Function if the seed is a static number (but of course generates the same random variables since the seed integer is the same). However, if I try to change it in each run (by 'From Workspace' block), it again gives error complaining about this matrix value not having a time value etc.
Do you have an advice to have different random variable realizations in Embedded MATLAB Function block in each parsim loop?
Thanks in advance,
Baris

  0 Comments

Sign in to comment.

Products


Release

R2017a

1 Answer

Answer by Rahul Kumar on 29 Aug 2018
 Accepted Answer

Here is an example to use different random seeds for each simulation
in(1:12) = Simulink.SimulationInput(mdlName);
for idx = 1:numWorkers
in(idx) = in(idx).setPreSimFcn(@(x) PreSimFcnCallback(idx));
end
function PreSimFcnCallback(seed)
rng(seed);
end
Please note that the example above is looping over 'numWorkers' the number of workers in the parallel pool, instead of the number of simulations, since it is sufficient to set the seed only once on each worker.

  0 Comments

Sign in to comment.