How to predict multiple time series data using deep learning

My dataset (18000 * 7 ) consists of air quality parameters. The parameters are PM10 concentration, wind speed, wind direction,temp, pressure, humidity, solar radiation. I want to predict the future pm10 concentration. I tried with the code available, but i cannot use it. please help me with the code

data=xlsread('pollutant.xlsx')

data=data';

dataTrainStandardized = (dataTrain - mu) / sig;

XTrain = dataTrainStandardized(1:end-1);

YTrain = dataTrainStandardized(2:end);

numFeatures = 6;

numResponses = 1;

numHiddenUnits = 200;

numTimeStepsTrain = floor(0.8*numel(data));

dataTrain = data(1:numTimeStepsTrain+1);

dataTest = data(numTimeStepsTrain+1:end);

mu = mean(dataTrain);

sig = std(dataTrain);

layers = [ ...

sequenceInputLayer(numFeatures)

lstmLayer(numHiddenUnits)

fullyConnectedLayer(numResponses)

regressionLayer];

options = trainingOptions('adam', ...

'MaxEpochs',250, ...

'GradientThreshold',1, ...

'InitialLearnRate',0.005, ...

'LearnRateSchedule','piecewise', ...

'LearnRateDropPeriod',125, ...

'LearnRateDropFactor',0.2, ...

'Verbose',0, ...

'Plots','training-progress');

dataTestStandardized = (dataTest - mu) / sig;

XTest = dataTestStandardized(1:end-1);

net = predictAndUpdateState(net,XTrain);

[net,YPred] = predictAndUpdateState(net,YTrain(end));

numTimeStepsTest = numel(XTest);

for i = 2:numTimeStepsTest

[net,YPred(:,i)] = predictAndUpdateState(net,YPred(:,i-1),'ExecutionEnvironment','cpu');

end

YPred = sig*YPred + mu;

YTest = dataTest(2:end);

rmse = sqrt(mean((YPred-YTest).^2))

figure

plot(dataTrain(1:end-1))

hold on

idx = numTimeStepsTrain:(numTimeStepsTrain+numTimeStepsTest);

plot(idx,[data(numTimeStepsTrain) YPred],'.-')

hold off

xlabel("Hour")

ylabel("Pm10 hourly concentration")

title("Forecast")

legend(["Observed" "Forecast"])

Answer 1

[Raunak Gupta]

Hi,

In my understanding you trying to use Regression based LSTM to predict the PM10 concentration as mentioned in the question. I assume that the PM10 concentration parameter resides in the last column of dataset file. I see you have used example from here. Here the number of features is only 1 that’s why you are facing problem.

The number of parameters you want to train on are 6, So I suggest changing YTrain in the code to have only the last column of the data. Also, the dataset split must be done to get all six parameters in the test and train data. Instead of using numel, try array indexing for getting appropriate number of samples for training and testing.

numTimeStepsTrain = floor(0.8*size(data,1));
dataTrain = data(1:numTimeStepsTrain+1,:);
dataTest = data(numTimeStepsTrain+1:end,:);

% DataSet Size --> (18000,7)
XTrain = dataTrainStandardized(:,1:6);
YTrain = dataTrainStandardized(:,7);
XTest = dataTestStandardized(:,1:6);
YTest = dataTestStandardized(:,7);

For multiple feature regression you can refer to the Sequence to Sequence Regression.

You can try out different other fine-tuning steps for getting much better results by looking into the following resources.

lstmLayer , trainingOptions , LSTM in MATLAB

[Athira t]

[file1,path1]=uigetfile('\*xlsx','pollutant concentration');

file1=fullfile(path1,file1);

data= xlsread(file1);

data= data';

numTimeStepsTrain = floor(0.8*size(data,1));

dataTrain = data(1:numTimeStepsTrain+1,:);

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

XTrain = dataTrain(:,1:6);

YTrain = dataTrain(:,7);

XTest = dataTest(:,1:6);

YTest = d(numTimeStepsTrain+2:end,7);

mu = mean(XTrain);

sig = std(XTrain);

mu1 = mean(YTrain);

sig1 = std(YTrain);

mu2 = mean(XTest);

sig2 = std(XTest);

mu3 = mean(YTest);

sig3 = std(YTest);

XTrain = (XTrain - mu) / sig;

YTrain = (YTrain - mu1) / sig1;

XTest= (XTest - mu2) / sig2;

YTest= (YTest - mu3) / sig3;

dataTrainStandardized = (dataTrain - mu) / sig;

XTrain = dataTrainStandardized(:,1:6);

YTrain = dataTrainStandardized(:,7);

XTest = dataTestStandardized(:,1:6);

YTest = dataTestStandardized(:,7);

still not working

[Raunak Gupta]

Hi,

You can look into the Sequence to Sequence Regression mentioned in the above answer as It contains end to end workflow of training the kind of Network you want. Also If you want to predict the future PM10 Concentration with previous PM10 concentration only then you can go ahead with the code you posted first with only the last column of the data matrix.

[Sharan Magavi]

as per my understanding you dont need to calculate mu and sig for all the variable ( XTrain, Test, etc)

you need to calculate it only on the Training dataset and APPLY the same mu and sig to the test set to normalise it.