You first need to create an input/output data base in order to train that network. I assume that calculation of the Heat transfer coefficient involves some kind of formula. If this is so, then you may program your network.
Considering the heat transfer coefficient as the output and the other variables as inputs (e.g., heat loss, difference in temperature, area of the surface, elapsed time), you could use the MLP, having some data based on this formula in order to train the network.