how can i test an time series NAR NN forecasting tool?

Dear all,

i would like to predicte the wind speed of a spesific location for the next 2 hours (12 points of 10min data) for that i am only using lagged input data of the time series (NAR AA tool).

i am trying to test the optimum number of delays (lagged inputs) needed to get a good prediction resutls using a for loop to increse the number of delays (for example between 1 and 10)

I am also trying to test the optimum number of inputs (input length) that need to be used for training the network and getting a better prediction. also by using another for loop (for example betweem number_of_delay_plus_one to length(input))

my code looks like the following:

if true
  clear
clc
load('C:\Users\00037218\Desktop\Wind Speed prediction\E82_Wind_1_2013.mat');
% Solve an Autoregression Time-Series Problem with a NAR Neural Network
% Script generated by NTSTOOL
% Created Fri Mar 28 22:27:25 CET 2014
%
% This script assumes this variable is defined:
%
%   Target_2 - feedback time series.
if 1 == 0
  prediction_indx = 12;
  time_input = Input_wind(:,1);
  time_target = (time_input(end)+mean(diff(time_input)):mean(diff(time_input)):time_input(end)+(prediction_indx*mean(diff(time_input))))';
  time = [time_input;time_target];
  Final = [time Final_test];
  i_end = length(Final_test);
  j_end = 10;
  for i = prediction_indx:prediction_indx:i_end
      targetSeries = tonndata(Final_test(1:i,:),false,false);
      targetSeriesVal = tonndata(Final_test(i+1:end,:),false,false);
      for j = 1:i-1%j_end
          feedbackDelays = 1:j;
          hiddenLayerSize = prediction_indx;
          net = narnet(feedbackDelays,hiddenLayerSize);

            % Choose Feedback Pre/Post-Processing Functions
            % Settings for feedback input are automatically applied to feedback output
            % For a list of all processing functions type: help nnprocess
            % TAKE CARE THIS LINE WAS ACTIVE: net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};

            % Prepare the Data for Training and Simulation
            % The function PREPARETS prepares timeseries data for a particular network,
            % shifting time by the minimum amount to fill input states and layer states.
            % Using PREPARETS allows you to keep your original time series data unchanged, while
            % easily customizing it for networks with differing numbers of delays, with
            % open loop or closed loop feedback modes.
            [inputs,inputStates,layerStates,targets] = preparets(net,{},{},targetSeries);

            % Setup Division of Data for Training, Validation, Testing
            % For a list of all data division functions type: help nndivide
            if 1 == 1
                net.divideFcn = 'dividerand';  % Divide data randomly
                net.divideMode = 'time';  % Divide up every value
                net.divideParam.trainRatio = 70/100;
                net.divideParam.valRatio = 10/100;
                net.divideParam.testRatio = 20/100;
            else
                net.divideFcn = 'divideblock';  % Divide data in blocks
                net.divideMode = 'time';  % Divide up every value
            end
            % Choose a Training Function
            % For a list of all training functions type: help nntrain
            net.trainFcn = 'trainlm';  % Levenberg-Marquardt

            % Choose a Performance Function
            % For a list of all performance functions type: help nnperformance
            net.performFcn = 'mse';  % Mean squared error

            % Choose Plot Functions
            % For a list of all plot functions type: help nnplot
            net.plotFcns = {'plotperform','plottrainstate','plotresponse', ...
                'ploterrcorr', 'plotinerrcorr'};

            % Train the Network
            [net,tr] = train(net,inputs,targets,inputStates,layerStates);

            % Test the Network
            outputs = net(inputs,inputStates,layerStates);
            errors = gsubtract(targets,outputs);
            performance = perform(net,targets,outputs)

            % Recalculate Training, Validation and Test Performance
            trainTargets = gmultiply(targets,tr.trainMask);
            valTargets = gmultiply(targets,tr.valMask);
            testTargets = gmultiply(targets,tr.testMask);
            trainPerformance = perform(net,trainTargets,outputs)
            valPerformance = perform(net,valTargets,outputs)
            testPerformance = perform(net,testTargets,outputs)

            % View the Network
            %view(net)

            % Plots
            % Uncomment these lines to enable various plots.
            %figure, plotperform(tr)
            %figure, plottrainstate(tr)
            %figure, plotresponse(targets,outputs)
            %figure, ploterrcorr(errors)
            %figure, plotinerrcorr(inputs,errors)

            % Closed Loop Network
            % Use this network to do multi-step prediction.
            % The function CLOSELOOP replaces the feedback input with a direct
            % connection from the outout layer.
            % ----------------------------------------------------------------
            delay=length(feedbackDelays); N=prediction_indx;
            targetSeriesPred = [targetSeries(end-delay+1:end), con2seq(nan(1,N))];
            % ----------------------------------------------------------------
            netc = closeloop(net);
            [xc,xic,aic,tc] = preparets(netc,{},{},targetSeriesPred);
            yc = netc(xc,xic,aic);
            perfc = perform(net,tc,yc);

            % Early Prediction Network
            % For some applications it helps to get the prediction a timestep early.
            % The original network returns predicted y(t+1) at the same time it is given y(t+1).
            % For some applications such as decision making, it would help to have predicted
            % y(t+1) once y(t) is available, but before the actual y(t+1) occurs.
            % The network can be made to return its output a timestep early by removing one delay
            % so that its minimal tap delay is now 0 instead of 1.  The new network returns the
            % same outputs as the original network, but outputs are shifted left one timestep.
            nets = removedelay(net);
            [xs,xis,ais,ts] = preparets(nets,{},{},targetSeries);
            ys = nets(xs,xis,ais);
            closedLoopPerformance = perform(net,tc,yc);
            figure();
            subplot(3,1,1);
            plot(cell2mat(inputs)); hold on;plot(cell2mat(outputs),'-.r');grid on; hold off;
            legend('Original Targets (Inputs)','Network Predictions (output)')
            subplot(3,1,2);
            plot(cell2mat(targets)); hold on;plot(cell2mat(yc),'-.r');grid on; hold off;
            legend('Original Targets','Network Predictions')
            subplot(3,1,3);
            plot(cell2mat(targetSeriesPred)); hold on;plot(cell2mat(yc),'-.r');grid on; hold off;
            legend('Original Targets_pred.','Network Predictions')
            save('Input_Data.mat','targetSeries','targetSeriesVal','i','j','prediction_indx')
            clear
            load('Input_Data.mat');
        end
    end
  end 
my question is:
1- what i am doing wrong:
   a- why my target data is changing even that i need it to be constant which mean i will always try to predicte the same next 2 hours with different delays used. (but i am not)
   b- how can i use the same input data length for training with changable 'for loop' number of lags (delays)?
   c- how to re-do the calcultion on a new input data length which is also changable 'for loop' in numaber

So simply speaking i am trying to do the following (only example):

if true
  for i = 5:5: length(inputdata)
    NN_input = inputdata(1:i);
    NN_target = inputdata(i+1:12);% 12 is for the next 2 hours it can be changed and it is 
% only used to compare
    forj = 1:1:10 % for changing the delays in the delay layer
    feedbackDelays = 1:j;
    * _% in this part my input and target data should stay the same *They are no*_ *
    % network code
    % stuff for testing the resutls
    % stuff for presenting the reustls

    end
  end

Thx in advance for you support

Answer 1

[Greg Heath]

Find feedback delays from the significant delays of the target autocorrelation function.

Find input delays from the significant delays of the input/target cross-correlation function.

Search the NEWSGROUP and ANSWERS using

 greg nncorr narnet

Hope this helps.

Thank you for formally accepting my answer

Greg

[Aubai]

Hi Greg,

I have already read your comments before asking my question and i know that the autocorrelation will help in obtaining the optimum number of delays (lags/ feedback) outputs. for that i have a question how can i get the significant delays? (is it the number of lags at which the autocorrelation coeficient at its peak?).

if so how can i get the optumum number of lags if i have two peaks at different lags (same value)?

another question is:

1- dose the partial autocorrelation help in such situation as stated before?

2- my real question is: why i am not getting the expected results from my code (by results i mean the same traget data with different lags /feedback delays/ as i am trying to use the same input for the NN and test for differet values of lags /delays/)?

as you can see my:

targetSeries = tonndata(Final_test(1:i,:),false,false);

targetSeriesVal = tonndata(Final_test(i+1:end,:),false,false);

my values are not changing in the second for loop so why am i seeing changes in the plots?

Thx in advance Aubai

[Greg Heath]

All of this is covered in previous NEWSGROUP and ANSWERS posts

 greg nncorr narnet

What have I done in my posts that you don't do in yours?

[Aubai]

hallo Greg,

thanks for your answer :)

I am trying to find the answer in those previous Answers, Newsgroup. but not fining it can you refer me to a spesific one?

another question: how the 'preparets' function works? i think my problem is there as iam getting random input data everytime the for-loop changes and this giving me not what i am expecting.

again i would like to: for the same available input data (for training) train the network for different delays number (changing with for-loop) and then predict the next 12 points in my time-series. so in my case i want the 'preparets' function to give me always the same set of input, validation and training data yet with different number of delays. how can i do this??

Best Regards