Unsupervised Learning for Lighting Control System

Finding a balance point between the energy consumption and user experience is one of the most important tasks in current lighting control systems. For achieving that goal, finding resonalble group configuration of sensors/luminaires is necessary. However, most of the current lighting control systems use fixed group configuration based on empirical knowledge from human experts. This thesis uses time-series clustering, a specific method in the unsupervised learning, for exploring possible solutions of building a model that can find that balance point automatically and dynamically. For evaluating the clutering result of our model, this thesis proposes three evaluation methods specifically designed for the lighting control system, since this topic is relatively new in this industry. After testing with difierent combinations of tools, this thesis builds a model that proved to be useful for solving the problem of finding proper group configurations for lighting control system