Notes

clc
clear

load('data20100513_online');
load('state1')
load('data20100210_online');
load('state2')
%% Step 1. Structure determination and initial parametr setting
n=3; %dimenze fuzzy systemu
M=10; %pocet pravidel 10 je ok
alpha=1; %learning rate (2 je nej s M=14 okno=100, q=1)
macBeta=1;
macGamma=0.1;
macN=-1;
X=do;
X1=Fm;
leBeta=2.59%2.52;
oknoM=20; %200 ok
%% Online initial parameter choosing method
pre_x=zeros(n,M);
pre_y=linspace(0,0,M+1);
for i=1:M
% u=i;
pre_x(2,i)=X(6+i*10);
pre_x(1,i)=X(i*10);
pre_x(3,i)=X(19+i*10); %X(19+i*10);
% pre_x(i)=X(i);
% pre_y(i)=X(i);
% pre_y(i+1)=pre_y(i)+sin(u);
end
%% Step 1. Structure determination and initial parametr setting
y=pre_y(2:M+1);
x=pre_x;

x;

sigma=rand([n,M])/.010;
pre_sigma=sigma;
for i=1:n
sigma(i,:)=(max(x(i,:))-min(x(i,:)))/(2*M);
end
x_data=linspace(0.5,1,n);
y_data=0.1;
ZZ=linspace(1,1,M);
z=zeros(n,M);

b=0;
a=0;
f=0;
vzorky=1600;
epoch=zeros(1,vzorky);
fuzzy_y=zeros(1,vzorky);
system_y=zeros(1,vzorky);
x_data=[0,0];
x;
f=0;
prirustek_y=zeros(vzorky,M);
prirustek=zeros(vzorky,M,2);
prirustek_sigma=zeros(vzorky,M,2);
err=zeros(vzorky,1);
for p=30:vzorky
y_data=X(p);
x_data(1)=X(p-1);
x_data(2)=X(p-6);
x_data(3)=X(p-19);

% if p==1500
% y_data=X(p)+0.01*X(p);
% X(p)=X(p)+0.01*X(p);
% x_data(1)=X(p-1);
% x_data(2)=X(p-18);
% end
% if (p<vzorky/2)
% x_data=y_data;%macBeta*(x_data)/(1+x_data^macN)-macGamma*x_data(k);
% % x_data(1)=sin(p/10);
% % x_data(2)=y_data;
% y_data=macBeta*(x_data)/(1+x_data^macN)-macGamma*x_data;
% x_data;
% else
% % x_data(1)=sin(p/10)*cos(x_data(1));
% % x_data(2)=y_data;
% % y_data=x_data(1)+x_data(2);
% % x_data;
% % x_data(1)=sin(p/10);
% % x_data(2)=y_data*0.5;
% % y_data=x_data(1)+x_data(2);
% % x_data;
% x_data=y_data;%macBeta*(x_data)/(1+x_data^macN)-macGamma*x_data(k);
% % x_data(1)=sin(p/10);
% % x_data(2)=y_data;
% y_data=macBeta*(x_data)/(1+x_data^macN)-macGamma*x_data(k);
% x_data;
% end
y_data=X(p);
x_data(1)=X(p-1);
x_data(2)=X(p-6);
x_data(3)=X(p-19);
%fuzzy_y(p)=f;
for q=1:10
%% Step 2. Present input and calculate the output of the fuzzy system
b=0;
a=0;
f=0;
ZZ=linspace(1,1,M);
z=zeros(n,M);
for l=1:M
for i=1:n
z(i,l)=membershipFunction(x_data(i),x(i,l),sigma(i,l));
ZZ(l)=ZZ(l)*z(i,l);
end

end
b=sum(ZZ);
for l=1:M
a=y(l)*ZZ(l)+a;
end
f=a/b;
if q==1,f-y_data;,end
err(q)=f-y_data;
if (p==13)
p;
end
if min(abs(err(q)))<0.0001,err(q)=err(q);break,end
y_old=y;
x_old=x;
sigma_old=sigma;
%% Step 3. Update the parameters.

for j=1:M % PICOVINY V INDEXOVANI
y(j)=y_old(j)-alpha*(f-y_data)*ZZ(j)/b;
prirustek_y(p,j)=abs(-alpha*(f-y_data)*ZZ(j)/b)+prirustek_y(p,j);
d_y(j,q)=-alpha*(f-y_data)*ZZ(j)/b;
for k=1:n
prirustek(p,j,k)=abs((alpha*(f-y_data)/b)*(y_old(j)-f)*ZZ(j)*2*(x_data(k)-x_old(k,j))/sigma_old(k,j)^2)+prirustek(p,j);
x(k,j)=x(k,j)-(alpha*(f-y_data)/b)*(y_old(j)-f)*ZZ(j)*2*(x_data(k)-x_old(k,j))/sigma_old(k,j)^2;
sigma(k,j)=abs(sigma_old(k,j)-(alpha*(f-y_data)/b)*(y_old(j)-f)*ZZ(j)*2*((x_data(k)-x_old(k,j))^2)/sigma_old(k,j)^3);
prirustek_sigma(p,j,k)=abs((alpha*(f-y_data)/b)*(y_old(j)-f)*ZZ(j)*2*((x_data(k)-x_old(k,j))^2)/sigma_old(k,j)^3)+prirustek_sigma(p,j);
end
end

end
epoch(p)=q;
system_y(p)=y_data;
fuzzy_y(p)=f;
end
max(prirustek)
%plot(err)
plot(system_y,'g')
hold on
plot(fuzzy_y,'r')
figure(2)
plot(epoch)
figure(3)
plot(sum(prirustek(:,:),2))
figure(4)
plot(sum(prirustek_sigma(:,:),2))
figure(5)
plot(sum(prirustek_y,2))
% for i=1:M
% figure(6)
% znormovano(i,:)=transpose(zscore(prirustek_y((1500+plovouci-okno+1):1500+plovouci,i)));
% plot(zscore(prirustek_y((1500+plovouci-okno+1):1500+plovouci,i)))
% hold on
% figure(7)
% plot(zscore(prirustek_sigma((1500+plovouci-okno+1):1500+plovouci,i)))
% end
% znormovano(znormovano<0)=0;
% bar(sum(znormovano))


%% LE
z_score_y=zeros(1800,M);
z_score_s=zeros(1800,M,n);
z_score_x=zeros(1800,M,n);

LE=zeros(1600,1);
for i=201:1600
for j=1:M
z_score_y(i,j)=(abs(prirustek_y(i,j))-mean(abs(prirustek_y(i-oknoM:i))))/std(abs(prirustek_y(i-oknoM:i)));
for k=1:n

z_score_s(i,j,k)=(abs(prirustek_sigma(i,j,k))-mean(abs(prirustek_sigma(i-oknoM:i,j,k))))/std(abs(prirustek_sigma(i-oknoM:i,j,k)));
z_score_x(i,j,k)=(abs(prirustek(i,j,k))-mean(abs(prirustek(i-oknoM:i,j,k))))/std(abs(prirustek(i-oknoM:i,j,k)));
end
end
z_score_y(z_score_y<0)=0;
z_score_s(z_score_s<0)=0;
z_score_x(z_score_x<0)=0;
z_score_s(isnan(z_score_s))=0;
z_score_x(isnan(z_score_x))=0;
z_score_x(isinf(z_score_x))=0;
z_score_s(isinf(z_score_s))=1000;
z_score_y(z_score_y<leBeta )=0;
LE_y(i)=sum(z_score_y(i,:));
LE_s1(i)=sum(z_score_s(i,:,1));
LE_s2(i)=sum(z_score_s(i,:,2));
LE_x1(i)=sum(z_score_x(i,:,1));
LE_x2(i)=sum(z_score_x(i,:,2));

end
figure(8)
plot(LE_y(250:1600))
% figure(9)
% plot(LE_s1(1000:2000))
% figure(10)
% plot(LE_s1(1000:2000)+LE_y(1000:2000))
% figure(11)
% plot(LE_x1(1000:2000))

toPlotLE=LE_y(1:1600);
toPlotLE(toPlotLE>100)=100;
toPlot_fuzzy=fuzzy_y((1:1600));
toPlot_sys=system_y(1:1600);
figure(9)
plot(toPlotLE(1:900),'LineWidth',2)
%title('Learning Entropy Order 1, beta=0, M=200','fontsize',14);
xlabel('k','fontsize',12)
ylabel('E^1 [-]','fontsize',12)
hold on
plot(1*state_20100210(1:900,2),'r','LineWidth',2)
% plot(1*state_20100513(1:900,2),'r','LineWidth',2)

figure(10)
plot(toPlot_fuzzy(30:930),'r','LineWidth',2)
%title 'time series vs adaptive predictor output'
xlabel('k','fontsize',12)
ylabel('y [%]','fontsize',12)
hold on
plot(toPlot_sys(30:930),'b','LineWidth',2)
hold off
figure(11)
plot(toPlot_fuzzy(30:930)-toPlot_sys(30:930),'LineWidth',2)
xlabel('k','fontsize',12)
ylabel('y [%]','fontsize',12)
title 'e [%]'