MATLAB: How to obtain a Simulink block for a neural network created in MATLAB code

deepgensimlearningnarxNetworkneuralnn

I have a NARX neural network which I have prepared and trained in MATLAB.

How do I obtain a Simulink block which behaves in the same way?

Best Answer

You can use the 'gensim' function to generate the Simulink block. Please find more information about 'gensim' on the following page:

https://www.mathworks.com/help/deeplearning/ref/gensim.html

You then need to set the neural network Simulink block initial conditions with 'setsiminit', as well as correctly configure the system. Please find more information about 'setsiminit' on the following page:

https://www.mathworks.com/help/deeplearning/ref/setsiminit.html

If you notice discrepancies between the network in MATLAB and in Simulink, please check that the initial values are set properly in the Simulnk block using 'getsiminit'.

Refer to the attached script for a simple example on how to do so.

Related Solutions

MATLAB: How to Create Simulink Block for NARX Neural Network and Predict Multiple Time Steps Ahead

Here is an analogous solution to a close loop neural network in Simulink that predicts multiple output steps, like the MATLAB example here:

https://www.mathworks.com/help/releases/R2018b/deeplearning/ref/narxnet.html#bvds6j8-25

I used the example from "setsiminit" to create this example for you, which is slightly modified:

% This part of the example is nearly identical
[x,t] = simplenarx_dataset;
xnew = x(81:100);
x = x(1:80);
t = t(1:80);
net = narxnet(1:2,1:2,20);
[Xs,Xi,Ai,Ts] = preparets(net,x,{},t);
net = train(net,Xs,Ts,Xi,Ai);
[Y,Xf,Af] = net(Xs,Xi,Ai);
[netc,Xic,Aic] = closeloop(net,Xf,Af);
% Show the predicted 20 steps in MATLAB
y2 = netc(xnew,Xic,Aic)
% Create a Simulink block
[sysName,netName] = gensim(netc,'InputMode','Workspace',...
'OutputMode','WorkSpace','SolverMode','Discrete');
setsiminit(sysName,netName,netc,Xic,Aic,1);
x1 = nndata2sim(x,1,1);

This will create a model in Simulink that outputs 20 values per Simulink time-step.

MATLAB: Is there a command to measure the training time of a NARX network

In order to monitor the training time or the epochs in the training of a NARX network, you can use the training record, which is the second argument returned by the train function. For example, if you have a NARX network in a variable "net", and network inputs and targets in variables "X", and "T" respectively, you can get the training record "tr" as follows:

>> [net,tr] = train(net,X,T);

The training record has various information about the training of the network which includes epochs in the training, and the training time. Please refer to the following link for more information on using the training record:

https://www.mathworks.com/help/nnet/ref/train.html#f8-670981

Also, if you need to specify maximum training time, or the number of epochs required for the training of your NARX network, "net", you can specify using the network training parameters. For example, you can access different training parameters for the network as follows:

>> net.trainParam

Further, to specify the maximum training time, or maximum epochs, you can set the "net.trainParam.time", or "net.trainParam.epochs" respectively.

Alternatively, you can measure the timing required for training by profiling the function using MATLAB Profiler, or by using the "tic" and "toc" stopwatch timer functions. Please refer to the following link for more information on profiling in MATLAB:

https://www.mathworks.com/help/matlab/matlab_prog/profiling-for-improving-performance.html

Best Answer

Related Solutions

MATLAB: How to Create Simulink Block for NARX Neural Network and Predict Multiple Time Steps Ahead

MATLAB: Is there a command to measure the training time of a NARX network

Related Question