Supervised Learning in Lighting Control Systems

The objective of the thesis is to develop Predictive Models for Lighting Control Systems. Lighting Systems typically employ various sensors to automatically control installed Luminaires. An example of such a sensor is the Passive Infrared (PIR), that detects human motion and subsequently changes the state of Luminaires. These sensors have pre-defined delay timer values which control the amount of activity in a lighting system. Luminaires are generally forced to stay on for a long period of time, to ensure that lights are not turned off when a room is occupied. However, using long delay timers also leads to excessive energy consumption. By developing predictive models, the system can anticipate human presence in advance, and tune light parameters to achieve the optimal balance. Using deep learning, it is shown that by framing the historical data of sensor outputs as time-series forecasting problem, it is possible to predict the output of a PIR sensor in advance, and use that information to develop lighting systems that can achieve higher energy savings than conventional solutions