A Simulation Study on the Performance of Machine Learning Daylight-Linked Lighting Control Under Urban Topography

Daylight-linked lighting control (DLLC) system has been recognized as one of the effective measures for improving indoor illuminance distribution and energy performance. However, the system is often considered ineffective due to poor calibration and commissioning of the system and lack of design guidelines (Bellia et al., 2016). Among different causes, the positions of photosensors play a crucial role in DLLC systems. The position affects the DLLC system in two major aspects. First, the location of the sensor needs to reflect the illuminance level on the working plan level where it would not be affected by a strong source of sunlight. Secondly, under normal circumstances, DLLC is controlled by a single photosensor which leads to over-dimming in the rear part of the room or overprovided electrical lighting at the front part of the room. Traditionally, the use of open-loop and closed-loop controls makes it impossible to change the dimming ratio of artificial lighting due to indoor illuminance distributions and climatic conditions. Hence, an artificial neural network (ANN)-based machine learning (ML) is used to identify the correlation between artificial lighting and photosensors under different operating conditions. This paper focuses on identifying the major input parameters for the ANN model. Findings show that the input parameters (features) have strong correlations with the dimming output. Also, the ANN model performed very well with very small errors in most of the cases.