MATLAB: Prediction of a time series

narxneural network

Hello,

I want to predict a time serious using the NARX-model of the Neural Network Toolbox. But in my special case the forecast function has to look like this:

y(t + 8) = f(y(t-4), y(t-5), … x(t-1), x(t-2), …)

At point t I know the value of y till y(t-4), the exegenous variables till point t-1.

How can I do this in Matlab?

Many thanks for all replies

Joerg

Best Answer

Think in terms of

y(t) = f(y(t-12), y(t-13), ... x(t-9), x(t-10), ...);

Hope this helps.

Thank you for formally accepting my answer.

Greg

Related Solutions

MATLAB: Change in fitness function

Faulty notation. Typical usage is input x, target t, output y . See the documentation examples for the regression/curve-fitting function FITNET.

See PATTERNNET for classification/pattern-recognition documentation examples.

Both functions call FEEDFORWARDNET which never has to be explicitly used.

 help fitnet
 doc fitnet
 [x, t] = simplefit_dataset;
 net    = fitnet(10);
 net    = train(net,x,t);
 view(net)
 y      = net(x);
 perf   = perform(net,t,y)      % Unscaled number doesn't tell you much

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% An expanded modification. Search the NEWSGROUP and ANSWERS using

 greg fitnet

% Ending semicolons removed from selected commands so that results are automatically printed to the screen

 [ x, t ] = simplefit_data; 
 [ I N ]  = size(x)
 [ O N ]  = size(t)
  figure(1) 
  plot(x,t) % Smooth curve with two local max and mins suggest at least 4 hidden nodes (H>=4)

% MATLAB Default trn/val/tst data division ratio is 0.7/0.15/0.15

Ntrn    = N-2*round(0.15*N)  % Default No. of training examples
Ntrneq  = Ntrn*O             % No of training equations
net     = fitnet;            % Uses default of one hidden layer with H = 10 hidden nodes
Nw      = (I+1)*H+(H+1)*O    % Nw = 31 unknown weights to estimate (Nw <= Ntrneq?)
 [ net tr y e ] = train(net,x,t);
 % y = net(x);
 % e = t-y;
 NMSE = mse(e)/var(t,1)        % Normalized mean-square-error  ( NMSE < 0.01 ?)
 R2   = 1 -NMSE                % Fraction of target variance modeled by the net (R2 > 0.99 ?)

Hope this helps.

Thank you for formally accepting my answer

Greg % See Wikipedia/Rsquared

MATLAB: How can i make a closed loop NarxNet with feedback of predicted values

 "but once predicted, feedback them"

That is exactly what narxnet does. The default values ID,FD = (1:2,1:2) yield

 1. y(t) = f( x(t-1), x(t-2), y(t-1), y(t-2)) for t >= 3

which can also be interpreted as

 2. y(t+3) = f( x(t+2), x(t+1), y(t+2), y(t+1)) for t>= 0

The nondefault input ID = 0:2 adds x(t) to the RHS of 1

The nondefault input FD = 0:2 adds y(t) to the RHS of 1 for OPENLOOP ONLY!!!

FD = 0 is not allowed for CLOSELOOP. Any attempt to do so will result in either an ERROR, or the command is ignored.

Hope this helps.

Thank you for formally accepting my answer

Greg

Best Answer

Related Solutions

MATLAB: Change in fitness function

MATLAB: How can i make a closed loop NarxNet with feedback of predicted values

Related Question