time series prediction with multiple series

Hi,

I want to predict the temperature of a room, where several heaters are located. The trained network will have input as heater powers as time series, and output as room temperature time series. I have a set of data with inputs and outputs that I can use to train the network. My goal is to use this network to predict temperature time series of the room for any given input time series heater powers.

To achive this goal, I have used NARX closed loop network but the accuracy was not enough (10-20degC error bandwith). I am new to MATLAB neural network tools, and I am not sure NARX is the correct one for this problem. Could you recommend any function for this goal? Any tutorial or example would be helpful.

Thanks!

Answer 1

[Ayush Anand]

I understand you want to do a timeseries prediction to predict the temperature of a room and are not getting enough accuracy with the NARX network. Re-initializing the network several times or tuning the parameters can help with an improved accuracy in many cases.

There are a few other MATLAB neural network tools that you can also try out for the same:

1. In some cases, a simple feedforward neural network can also be effective for time series prediction tasks. You can try building a feedforward network using the "feedforwardnet" function in MATLAB.
2. Long Short-Term Memory Network (LSTM): LSTM networks are networks are well-suited for modelling sequential data and have been successful in various time series prediction tasks. MATLAB provides the "lstmLayer" and "trainNetwork" functions to build and train LSTM networks.

For more examples on how to use MATLAB neural network tools for timeseries prediction, you can type "nnstart" in the MATLAB command window and select the "Neural Network Time Series" category.

I hope this helps!