The neural network house: An overview

Typical home comfort systems utilize only rudimentary forms of energy management and conservation. The most sophisticated technology in common use today is an automatic setback thermostat. Tremendous potential remains for improving the efficiency of electric and gas usage. However, home residents who are ignorant of the physics of energy utilization cannot design environmental control strategies, but neither can energy management experts who are ignorant of the behavior patterns of the inhabitants. Adaptive control seems the only alternative. We have begun building an adaptive control system that can infer appropriate rules of operation for home comfort systems based on the lifestyle of the inhabitants and energy conservation goals. Recent research has demonstrated the potential of neural networks for intelligent control. We are constructing a prototype control system in an actual residence using neural network reinforcement learning and prediction techniques. The residence is equipped with sensors to provide information about environmental conditions (e.g., temperatures, ambient lighting level, sound and motion in each room) and actuators to control the gas furnace, electric space heaters, gas hot water heater, lighting, motorized blinds, ceiling fans, and dampers in the heating ducts. This paper presents an overview of the project as it now stands.

Retrieve Paper (postscript)
Retrieve Paper (pdf)