rgnn_sim.m

介绍如何用较复杂的方法去优化对负载具有感知特性的“黑匣子”模型的神经网络控制。

function [Y_hat x]=rgnn_sim(U,W,activation,x,n)
%[Y_hat x]=rgnn_sim(U,W,activation,x,n)
%this calculates feedback gnn with selectable activation function
%U=input (one column is one time step) The last row should be ones
%W=weights
%activation=vector of activation functions:
%	0=sigmoid
%	1=tanh
%	2=linear
%	3=bipolar squash (x/(abs(x)+1))
%n=number of outputs

%    Copyright Travis Wiens 2008
%
%    This program is free software: you can redistribute it and/or modify
%    it under the terms of the GNU General Public License as published by
%    the Free Software Foundation, either version 3 of the License, or
%    (at your option) any later version.
%
%    This program is distributed in the hope that it will be useful,
%    but WITHOUT ANY WARRANTY; without even the implied warranty of
%    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
%    GNU General Public License for more details.
%
%    You should have received a copy of the GNU General Public License
%    along with this program.  If not, see <http://www.gnu.org/licenses/>.
%
%    If you would like to request a commerical (or other) license, please
%    feel free to contact [e-mail]

if nargin<5
    n=1;%default to one output
end

if nargin<4
    x=zeros(size(W,1));
end

if nargin<3
    activation=ones(size(W,1),1);%default to tanh
    activation(end)=2;%linear output
end

if n~=1
	error('This function currently only works with single outputs (n=1)')
end

[N_t1 N_t2]=size(W);
if N_t1~=N_t2
    error('W must be square')
end
N_t=N_t1;
[m K]=size(U);%number of inputs, number of time steps
N=N_t1-n;%number of input + hidden neurons
x=zeros(N+n,K);%initialize neuron outputs
Y_hat=zeros(n,K);%initialize network outputs

for k=1:K;%loop through time steps
    if k>1
	x(:,k)=x(:,k-1);
    end
    x(1:m,k)=U(:,k);%copy inputs to net

    for i=m+1:N_t;%loop through neurons
        net=0;
        for j=1:N_t;%calculate wieghted sum
            net=net+W(i,j)*x(j,k);
        end
        if activation(i)==0;%%sigmoid
          x(i,k)=1/(1+exp(-net));
	elseif activation(i)==1;%tanh
	  x(i,k)=tanh(net);
	elseif activation(i)==2;%linear
	  x(i,k)=net;
	elseif activation(i)==3;%bipolar squash
	  x(i,k)=net/(abs(net)+1);
        else
          error('Unknown Activation function')
	end
    end
    Y_hat(:,k)=x(N+1:n+N,k);%output
end