Intelligent lighting systems: Leveraging LSTM and time-series clustering for optimal PhotoSensor deployment

Optimizing artificial lighting in indoor spaces is a critical strategy for reducing energy consumption, enhancing visual comfort, and promoting environmental sustainability. One of the key challenges in intelligent lighting control lies in determining the optimal placement of photosensors. This study presents a methodology that leverages Long Short-Term Memory (LSTM) neural networks to predict workplane illuminance from data collected by multiple photosensors. The primary objective is to identify the sensor position that exhibits the strongest correlation with workplane illuminance. Acknowledging that several photosensors may yield comparable correlations, the study further incorporates a time-series clustering analysis to evaluate the temporal patterns of daylight illuminance across photosensors. This approach further refines sensor selection by analysing temporal trajectories of daylight illuminance that best support stable lighting control and reduce adaptive oscillations. The methodology was tested in two laboratory environments with distinct spatial and lighting configurations. In the first case study, the photosensor placed on a wall outperformed those on the ceiling, indicating a clearly optimal placement. In the second case study, all sensors showed similar high predictive performance, justifying the use of time-series clustering to capture temporal illuminance patterns to refine selection. These results confirm the effectiveness of the proposed methodology in guiding sensor placement decisions for advanced lighting control systems.