Reinforcement Learning Model of Irregular Shading Control for Optimal Daylight and Glare Analysis

Daylight plays an important role in building environments. An environment with better indoor daylight can improve occupants’ health and productivity. Shading devices are often used to avoid visual discomfort while maximizing daylight availability. The most used shading device nowadays are blinds with unified slat angle control in regular shapes. With trend towards uniqueness in high-rise buildings, shading devices with irregular shapes are becoming popular and more and more commonly used. However, the traditional method of finding an optimal control strategy is unfit for dynamic irregular shape shading devices. Compared with regular shape shading devices, irregular shape shading devices have various rotation angles, and each component can be independently controlled. Therefore, irregular shading devices require the collaboration of multiple components to control indoor daylight. However, the complex combinations of different angles and different components make it impossible for the traditional method to find the optimal control strategy. To address this issue , this paper proposes a selflearning shading control algorithm based on reinforcement learning trained towards optimal working surfaces throughout the day. The irregular shading device will learn the optimal angle based on discrepancies between the desired and the current indoor visual environment quality.