Modelling of two complex heating, ventilating, and air conditioning (HVAC) relationships using neural networks is reported. Such networks have the potential for exact representations, without approximations, and also possess a structure that facilitates hardware implementation. The two applications considered are the estimation of the predicted mean vote (PMV) for thermal comfort as defined by Fanger:P:O., and the generation of the psychrometric chart. Calculation of the PMV index is performed using primary and correction subcalculations as suggested by Fanger. The proposed network architecture takes advantage of Fanger's estimation technique. The other application involves psychrometric calculations using neural networks. Both architectures give very good results and are potentially attractive for real-time control applications.

KEYWORDS: building services, calculating, expert systems, thermal comfort, psychrometry, relative humidity, accuracy.

Units: SI