Machine Learning Approach For Building Management Systems

As buildings strive to improve energy efficiency, productivity, and overall well-being, the use of advanced sensing and control systems is becoming increasingly common. Unlike reactive control systems, these systems are more sophisticated and do not rely on fixed setpoints and rigid schedules, allowing for greater building performance. To address the complex challenges of building control, modern machine learning techniques are utilized. In particular, agent-based models have proven effective in managing intricate technical systems such as HVAC systems. This approach involves autonomous agents that constantly adjust and make decisions in response to changing conditions. Although numerous industries have extensively studied adaptive agents in ABMs, limited attention has been given to the real-world implementation of AI optimization techniques. However, traditional optimization methods such as mixed integer linear prediction and gradient descent typically yield the most effective solutions. This paper delves into exploring a model-free, deterministic, actor-critic, gradient-based continuous control approach to achieving the desired supply air temperature. The proposed approach involves a thermal energy storage agent that determines the ideal valve position for regulating the temperature of the cooling water flow as a component of BMS.

Keywords:

BMS , Building Management , Machine Learning , AI , BMS Applications