my prediction for multi step ahead prediction not giving good result as i expect (attached output as pdf)

i have simulated my data with the following codes. but seems like it is giving wrong answer. how a prediction gives like a parralal line. my data set Y=[182 325 390 480 519 620 630 609 635 700 660 615 561.15 534 589.9 703.45 651.02 674.62 661.42 652.66 646.34 642.4 664.34 660.03 683.11 664.66 688.61 697.53 693.71 680.43 751.3 ]

plt = 0; 
% T = simplenar_dataset;
N = length(Y)                   % 100
net = narnet( 1:2,[] );              % default inputs
view(net)
T = tonndata(Y,false,false);
[ Xs, Xsi, Asi, Ts ] = preparets( net, {}, {}, T );
ts = cell2mat( Ts );
plt = plt+1; 
figure(plt), hold on
plot( 3:N, ts, 'LineWidth', 2 )
 rng( 'default' )
[ net tr Ys Es Af Xf ] = train( net, Xs, Ts, Xsi, Asi );
% [ Ys Xf Af ] = net( Xs, Xsi, Asi )
% Es = gsubtract( Ts, Ys )
view( net )
NMSEs = mse( Es ) /var( ts,1 )          % 1.8107e-08
ys = cell2mat( Ys );
plot( 3:N, ys, 'ro', 'LineWidth', 2 )
axis( [ 0 31 180 800 ] )

legend( 'TARGET', 'OUTPUT' )
title( 'OPENLOOP NARNET RESULTS' )
                % 7. Now, the output feedback loop is closed and tested so that predictions can 
% eventually be extended beyond the time of the known target, T. 

   [ netc Xci Aci ] = closeloop(net,Xsi,Asi);
   view(netc)
   [Xc,Xci,Aci,Tc] = preparets(netc,{},{},T);
   [ Yc Xcf Acf ] = netc(Xc,Xci,Aci);
   Ec = gsubtract(Tc,Yc);
   yc = cell2mat(Yc);
   tc = ts;
   NMSEc = mse(Ec) /var(tc,1) 

%3.505e-07 
% Although the performance is a factor of 20 worse, it is still excellent. If it were  significantly degraded, %netc would be trained starting with the weights obtained  in the openloop stage (NOTE: This is not %mentioned in either %the help or doc documentation). 
% 8. The removedelay function is included in both documentations. However, I  see no good reason for %using it here. For example, when used with the default  output feedback delay 1:2, it yields invalid zero %and/or negative %values. 
%IS THIS A BUG THAT NEEDS TO BE FIXED ?
% 9. To extend the prediction beyond the end of the known series, empty cells are 
% used as inputs and the final delay states Xcf and Acf are used as initial conditions. and are used
     % [ Yc Xcf Acf ] = netc( Xc, Xci, Aci );
 Xc2 = cell(1,N);
[Yc2 Xcf2 Acf2] = netc( Xc2, Xcf, Acf ); 
yc2 = cell2mat(Yc2);
plt = plt+1; figure(plt), hold on
plot( 3:N, tc, 'LineWidth', 2 )
plot( 3:N, yc, 'ro', 'LineWidth', 2 )
plot( N+1:2*N, yc2, 'o', 'LineWidth', 2 )
plot( N+1:2*N, yc2, 'r', 'LineWidth', 2 )
axis( [ 0 N+15 180 900 ] )
legend( 'TARGET','OUTPUT','PREDICTION ')
title( 'CLOSED LOOP NARNET RESULTS ' )

Answer 1

[Greg Heath]

net = narnet( 1:2, [] ); % default inputs

GEH1: Error; H=[] means H=0, whereas the MATLAB default is H=10

T = tonndata(Y,false,false);

GEH2: Error; THIS CAUSES ERROR in preparets. If Y is a double, the second input should be true

 [ net tr Ys Es Af Xf ] = train( net, Xs, Ts, Xsi, Asi );

GEH3: Error; = Interchange Af and Xf

NMSEs = mse( Es ) /var( ts,1 ) % 0.2943

GEH4: = Too high. Want < 0.005 (Probably caused by H=[];

Accordingly, NMSEc = 0.8707 is unacceptable.

Training netc to lower NMSEc probably will not help because H=0;

Hope this helps.

Thank you for formally accepting my answer

Greg