Privacy-aware Fully Model-Free Event-triggered Cloud-based HVAC Control

Abstract

Privacy is a major concern when computing-as-a-service (CaaS) platforms, e.g., cloud-computing platforms, are utilized for building automation, as CaaS platforms can infer sensitive information, such as occupancy, using the sensor measurements of a building. Although the existing encrypted model-based control algorithms can ensure the security and privacy of sensor measurements, they are highly complex to implement and require high computational resources, which result in a high cost of using CaaS platforms. To address these issues, in this paper, we propose an encrypted fully model-free event-triggered cloud-based HVAC control framework that ensures the privacy of occupancy information and minimizes the communication and computation overhead associated with encrypted HVAC control. To this end, we first develop a model-free controller for regulating indoor temperature and CO2 levels. We then design a model-free event-triggering unit which reduces the communication and computation costs of encrypted HVAC control using an optimal triggering policy. Finally, we evaluate the performance of the proposed encrypted fully model-free event-triggered cloud-based HVAC control framework using the TRNSYS simulator, comparing it to an encrypted model-based event-triggered control framework, which uses model predictive control to regulate the indoor climate. Our numerical results demonstrate that, compared to the encrypted model-based method, the proposed fully model-free framework improves the control performance while reducing the communication and computation costs. More specifically, it reduces the communication between the system and the CaaS platform by 64% amount, and its computation time is 75% less than that of the model-based control.