gpn_ada.m.windows

this is matlab code for orthogonal freqency

% Simulate a goal programming network
% Adaptive learning rate strategy has been used (cfr SuperSAB)
%
% Usage: [V,F,E,U,M,conv_flag] = gpn_ada (D, B, G, ep, dt, U_init, m_init)
%
%		V	Final Network State
%		F	Constraint Satisfaction (F outputs)
%		E	sum(abs(F))
%		D,B	Constraint arrays
%		G	Gain Vector [nc x 1]
%		M	Learning rate trace

function [V,F,E,U,M,conv_flag] = gpn (D, B, G, ep, dt, U_init, m_init)

if (nargin<2)
	ep = 1000;
	dt = .001;
end

[nc ns] = size(D);	% # states, # constraints
conv_flag = 0;

	%%%%%%%%%%%%%%%%%%%%%%%%%
	% Array Initialisations %
	%%%%%%%%%%%%%%%%%%%%%%%%%
U = zeros (ep,ns);
if (exist('U_init'))
	U(1,:) = U_init;
end
E = zeros (ep-1, 1);
PF = zeros (ep-1,nc);
M = zeros (ep-1,1);

	%%%%%%%%%%%%%
	% Here Goes %
	%%%%%%%%%%%%%
m = m_init;
e = 2;
	% Compute Network States g(u_i)
	V = U(e-1,:);

	% Output F amps
	F = G.*(sum(D.*(ones(nc,1)*V),2)-B);
	E(e-1) = sum(abs(F));

	% Update Network States
	aux2 = -sum(D.*(F*ones(1,ns)),1);
	U(e,:) = U(e-1,:) + (m*dt).*(aux2);

	PF(e-1,:) = F';
	M(e-1) = m;

for e=3:ep
	% Compute Network States g(u_i)
	V = U(e-1,:);

	% Output F amps
	F = G.*(sum(D.*(ones(nc,1)*V),2)-B);
	E(e-1) = sum(abs(F));

	% Adaptive Learning Rate
	if (E(e-1)==0)
		conv_flag = 1;
		break;
	end
	if (E(e-2)/E(e-1)>.9999)
		m = m*1.05;

		% Update Network States
		aux2 = -sum(D.*(F*ones(1,ns)),1);
		U(e,:) = U(e-1,:) + (m*dt).*(aux2);
	else
		m = m/2;
		U(e,:) = U(e-2,:);
	end

	PF(e-1,:) = F';
	M(e-1) = m;
end