Towards Machine Learning-based Model Predictive Control for HVAC Control in Multi-Context Buildings at Scale via Ensemble Learning

This paper proposes a new framework to provide the ML forecasting model for model predictive control (MPC) in building HVAC systems. Buildings typically encompass multiple contexts, such as different types of rooms, each with distinct requirements for the ML models used in MPC. However, developing customized models requires significant effort. The proposed solution addresses this challenge through ensemble learning techniques, which involve grouping a set of existing pre-trained models to construct a new model tailored to the target context. This work employs a Bayesian Optimization algorithm, with the pre-trained models supported by an established AI platform in the building sector. On-site experimental results from two case studies demonstrate that the proposed solution reduces energy consumption by 7.96 kWh (52.4%) compared to using a single forecasting model.