Negative values in NARX NN for multi-step ahead prediction

i am getting negative values in performance (MAPE,MSE etc) as shown in attached figure in multi-step ahead prediction.the trained network performs well on training data and giving awkward results on Test data.My script is:

%% 1. Importing data
 
clc
clear all

 data3=xlsread ('HomeA-meter_2016.xlsx','C:C');
 %% 
 
 data=data3(1:8000)';
 %%

% Partition the training and test data. Train on the first 75% of the sequence 
% and test on the last 25%
numTimeStepsTrain = floor(0.75*numel(data));

dataTrain = data(1:numTimeStepsTrain+1);
dataTest = data(numTimeStepsTrain+1:end);

 %% Standardize Data 
% For a better fit and to prevent the training from diverging, standardize the 
% training data to have zero mean and unit variance. 
mu = mean(dataTrain);
sig = std(dataTrain);

dataTrainStandardized = (dataTrain - mu) / sig;
dataTestStandardized = (dataTest - mu) / sig; 
 
% N=300;

XTrain = dataTrainStandardized(1:end-1);
YTrain = dataTrainStandardized(2:end);


dataTestStandardized = (dataTest - mu) / sig;

XTest = dataTestStandardized(1:end-1);
YTest = dataTest(2:end);

% Training data 
  X = tonndata(XTrain,true,false);
  T = tonndata(YTrain ,true,false);
% Test data 
  X2 = tonndata(XTest,true,false);
  T2 = tonndata(YTest ,true,false); 
% data=normalize(data2);

% 
%% 


%% 2. Data preparation

inputSeries  = X;
targetSeries =  T;
% 2nd group: this is the new data used for testing trained network. 
inputSeriesVal  = X2;
targetSeriesVal =  T2 ; % This is generally not available
%% 3. Network Architecture
delay = 2;
neuronsHiddenLayer = 20;

% Network Creation
net = narxnet(1:delay,1:delay,neuronsHiddenLayer);
%% 4. Training the network
[Xs,Xi,Ai,Ts] = preparets(net,inputSeries,{},targetSeries); 

 net.divideFcn = 'dividetrain'; % Divide data in blocks 
 net.divideMode = 'time'; % Divide up every value
% % net.divideParam.trainRatio = 75/100;
% % net.divideParam.valRatio = 25/100;
% % net.divideParam.testRatio = 00/100;
  net.trainParam.epochs=1000;


net = train(net,Xs,Ts,Xi,Ai);
% view(net)
Y = net(Xs,Xi,Ai); 
% Performance for the series-parallel implementation, only 
% one-step-ahead prediction
perf = perform(net,Ts,Y);

%% Prediction on Training data

figure(1)

error=(cell2mat(Ts)-cell2mat(Y));

rmse = sqrt((mean(error).^2))
MAPE_Training=mean(abs(cell2mat(Ts)-cell2mat(Y))./cell2mat(Ts))*100


%% 5. Multi-step ahead prediction
[Xs1,Xio,Aio] = preparets(net,inputSeries(1:end-delay),{},targetSeries(1:end-delay));
[Y1,Xfo,Afo] = net(Xs1,Xio,Aio);
[netc,Xic,Aic] = closeloop(net,Xfo,Afo);
[yPred,Xfc,Afc] = netc(inputSeriesVal,Xic,Aic);
multiStepPerformance = perform(net,yPred,targetSeriesVal);

%% Prediction on Test data

 YTest=cell2mat(targetSeriesVal);
 Y2Pred=cell2mat(yPred);
 
 % Unstandardize the predictions using the parameters calculated earlier.
YPred = sig*Y2Pred + mu;

error2=(YTest-YPred);

rmse2 = sqrt((mean(error2).^2))
 
MAPE_Test=mean(abs(YTest-YPred)./YTest)*100


figure(2)


plot(cell2mat(inputSeriesVal),'linewidth',1);
hold on
plot(cell2mat(yPred),'linewidth',1)
xlabel("Date and Time (Number of samples)")
ylabel("EDemand ")
title("Prediction on Test data (XTest,YTest)")
legend(["Real data" "NARX prediction (Testing) "])
 xlim([0 100])
 hold on

%%  
 YTest=cell2mat(targetSeriesVal);
% 
 Y2Pred=cell2mat(yPred);
  % Unstandardize the predictions using the parameters calculated earlier.
 YPred = sig*Y2Pred + mu;
% 
%  rmse = sqrt(mean((YTest-YPred).^2))
 
 
 
numTimeStepsTest = numel(XTest);

%% Future prediction

% Plot the training time series with the forecasted values.

figure (4)
plot(dataTrain(1:end-1))
hold on
idx = numTimeStepsTrain:(numTimeStepsTrain+numTimeStepsTest);
plot(idx,[data(numTimeStepsTrain) YPred],'r' , 'linewidth' , 2)
% plot(idx,[data(numTimeStepsTest) YPred],'r' , 'linewidth' , 2) % Ali
% xlim([375000 385000]) Home C
%  xlim([175000 195000]) %Home B
% xlim([183000 186000]) %Home B
 xlim([5900 6100]) %Home B
hold off
xlabel("Date and Time (Number of samples)")
ylabel("Demand ")
title("Forecast future steps")
legend(["Real data" "Future prediction "])

% 
 rmse_test = sqrt(mean((YTest-YPred).^2))
 MAPE_Test=mean(abs(YTest-YPred)./YTest)*100
 
%%