Indoor haze particulate control using knowledge graphs within self-optimizing HVAC control systems

Transboundary haze pollution in South East Asia is posing a threat to conventional design of buildings yet indoor air pollution from haze particulate infiltration has still received less attention in Malaysia compared to haze pollution outdoors. Because of this minimal research effort, indoor building environments have increasingly become very complex environments for facility managers to monitor and control due to the corresponding growth in heterogeneity within building behavioural information and monitoring (sensory) systems. As a solution to this and part of an ongoing study, this paper presents the preliminary process of modelling heterogeneous building information related to indoor air quality (IAQ) (building envelope, sensors, contaminant properties, geometry, occupancy schedules and weather data) within modular and extensible semantic web knowledge graphs (KG). This work argues that this data model can preserve the existential and latent parametric relationships within such information therafter availing an accurate representation of the heterogeneous state-space in machine learning workflows of self-learning building monitors and controllers. Compared to the conventional homogenous feature vectors, KGs hold sufficient context-aware semantics for an algorithmic building control system to smartly monitor the IAQ and autonomously learn to adapt air handling units towards occupant comfort in an energy efficient manner. Specifically, this paper highlights the high-level implementation process of KGs within the deep Q-learning process of the aforementioned control systems. Finally, a brief discussion is provided on how this process reduces the complexity that facility managers face while operating their IAQ control systems followed by the conclusions and future work to be carried out in this study.