Beginning in the early 1980s, diagnostic approaches for HVAC systems in commercial and institutional buildings have been developed using progressively more sophisticated methods. They have met with some success but have not been widely and eagerly adopted in the marketplace. A new, statistically rigorous method was recently developed for whole building, energy-based problem detection. In this new approach, neural networks are organized into a higher-level model called a belief network, which can be viewed as a probabilistic database containing what is known about a system (Pearl 1988). The whole-building energy (WBE) module described here is one module of a larger system for whole building diagnostics developed by a team of private sector, national laboratory, and university researchers. This paper reports on the new approach, the algorithms used, and a test of the software on a data set recorded in a commercial building.

Units: I-P