Considers the real-world problem of analysis of the nonlinear relationships among the factors affecting building thermal loads in addition to the modelling problem itself. Steam and electricity requirements for two on-campus buildings have been dynamically modelled using nonlinear neural networks. Published National Weather Service data for the site provide the weather inputs. The neural model captures the dynamic relationships among 17 relevant inputs and the corresponding thermal load outputs well. After validation via statistical analysis, the neural models are used in a novel sensitivity analytical technique to determine the relative contribution of four weather-related factors affecting these thermal loads. Results show that ambient temperature has a huge impact on building steam consumption, while relative humidity has the largest effect on building electric load. Such sensitivity analyses reveal valuable information about thermal behaviour of buildings that can be effectively used for building thermal design. This technique also has considerable potential for applicability to a large class of HVAC systems.

KEYWORDS: year 1997, calculating, heat load, buildings, load, expert systems, electricity consumption, energy consumption, steam, testing, comparing, weather, solar radiation