Development of microservices with machine learning algorithms for natural ventilation control in smart buildings

Abstract

Smart buildings often struggle with the automatic control of complex heating, ventilation, and air conditioning systems, especially natural ventilation control. This paper introduces a novel microservices architecture to enable machine learning (ML) algorithms for natural ventilation control experiments in smart buildings. Implemented and evaluated in a three-story smart building in Cambridge, MA, from 2019 to 2021, the architecture incorporates a Python-based IoT network API and a Weather Forecast API. Experimental research demonstrated that predictive and reinforcement learning algorithms effectively controlled natural ventilation, optimizing CO2 levels (800–900 ppm) and indoor air temperature (below 26 °C). Additionally, augmented TABS control, leveraging solar radiation prediction, successfully prevented overheating and saved heating energy. This study highlights the critical importance of microservices architecture in transforming complex building systems into scalable, resilient IoT frameworks for control research, enabling advanced ML for more climate-responsive and energy-efficient buildings.