Good afternoon, I am trying to use the Neural Networks Time Series Toolbox in order to model the relationship between Solar Radiation, External Temperature and Heating Request of a building and its Internal Temperature. I would like to use a network with backpropagation such that I can simulate the internal temperature at timestep "t+1" by using the data of the internal temperature at timestep "t" previously calculated by the network, together with other input data. I would like to obtain this: timestep t : INPUT DATA=[Solar Radiation(t), External Temperature(t), Heating Demand(t)] + first attempt of the internal temperature
timestep t+1: INPUT DATA=[Solar Rad(t+1),External temperature(t+1), Heating Demand(t+1)] and internal temperature calculated by the network at timestep t (backpropagation).
and so on..
For this reason I used the NARX kind of network, but after creating the network, when I simulate it, it seems that in the "inputs" cell array the whole target result vector is already contained. [% Test the Network outputs = net(inputs,inputStates,layerStates);] The cellarray I have to feed to the newtork is like:
[3x1 double] => with SolRad,ExtTemp and HeatDem
[Internal Temperature target]
But how can I feed the network with the internal temperature if it is what I am trying to calculate??
The question is: -Is the network simulating the whole day at once? If yes, how can I simulate a timestep at once? -Is the backpropagation working? -Should I use another kind of neural network in order to do this?
Thanks is advance to anyone that will help, I hope my problem is clear and I am available for any further explanation.
SD
