Architecture for intelligent thermostats that learn from occupants' behaviour.

This paper proposes a new approach to thermostat design. For many years, thermostats have been "dumb" devices, meaning that they react to their environment either by direct user control or by previous user programming. This new approach details an intelligent thermostat that learns about the behavior of the occupants and their environment and controls ambient temperature to maintain comfort according to human specifications. In that way, the thermostat reduces the number of interactions with the user and eliminates the need for them to learn how to program the device. Additionally, the thermostat reduces energy consumption by setback when occupants are absent. While the proposed architecture fundamentally changes the functionality of today's conventional thermostats, it retains their simple user interface. This article presents the modular software architecture of this new intelligent thermostat design. The functionality of the thermostat in different states is described and how each module specializes in learning a certain pattern is explained. At the end, the results obtained using neural networks as a technique for learning are presented.