Create custom NARX net

Hi,

I'm strugglinng to create a series parallel architecture net(Pic1). I want to use this architecture to train my net.

Could somebody tell me how I can connect the ouptut to the first layer? Aferwards I'd like to use this net:

CODE (Pic2)

    BattCurrent = Experiment.Results(1).BattCurrent__A_;
    CellVolt = Experiment.Results(1).CellVolt__V_;
    SOC = Experiment.Results(1).SOC__0_1_;
    CellTemperature = Experiment.Results(1).CellTemperature__K_;
    NumberOfChargeProcedures = Experiment.Results(1).NumberOfChargeProcedures____;
    AgeingCapacity = Experiment.Results(1).AgeingCapacity;

    % Input Vektor X
    X = [BattCurrent CellVolt SOC CellTemperature NumberOfChargeProcedures]';
    %X = con2seq(X);

    %Output Vektor T
    T = [AgeingCapacity]';
    %T = con2seq(T);

    [Xn,Xs] = mapminmax(X);
    [Tn,Ts] = mapminmax(T);

    % ANN

    net = network;
    net.name = 'Test';
    net.numInputs = 1;
    net.numLayers = 3;

    net.biasConnect = [1; 1; 1];
    net.inputConnect(1,1) = 1;
    net.layerConnect(2,1) =1;
    net.layerConnect(3,2) =1;
    net.layerConnect(1,3) =1;
    net.outputConnect(1,3) = 1;

    %Layers
    net.layers{1}.size = 15;
    net.layers{1}.transferFcn = 'tansig';
    net.layers{1}.initFcn = 'initnw';
    net.layers{1}.name = 'Hidden Layer 1';

    net.layers{2}.size = 15;
    net.layers{2}.transferFcn = 'tansig';
    net.layers{2}.initFcn = 'initnw';
    net.layers{2}.name = 'Hidden Layer 2';

    net.layers{3}.size = 1;
    net.layers{3}.transferFcn = 'purelin';
    net.layers{3}.initFcn = 'initnw';
    net.layers{3}.name = 'Output';

    %NARX
    net.layerWeights{1,3}.delays = [1];

    %Functions
    net.initFcn = 'initlay';
    net.performFcn = 'mse';
    net.trainFcn = 'trainbr';
    net.divideFcn = 'dividerand';

    %Plots
    net.plotFcns = {'plotperform','plottrainstate'};

    view(net)

Thank you in advance! Best, Stefan

Answer 1

[Jayaram Theegala]

You can use "closeloop" function to connect output to the first layer, in other words to convert a neural network into a closed loop network. For more information about this function, click on the following URL:

https://www.mathworks.com/help/nnet/ref/closeloop.html

After creating the above closed loop network, you can create a feed forward network using the "feedforwardnet" function, and to find more information about this function click on the following MATLAB documentation page:

https://www.mathworks.com/help/nnet/ref/feedforwardnet.html

Answer 2

[Greg Heath]

 See the documentation examples

 help narxnet
 doc  narxnet

The only significant difference between your design and the documentation examples is that you have 2 hidden layers

 However 
 1. Use DIVIDEBLOCK  for training

Hope this helps.

Greg