Air conditioning (AC) plays a major role in building energy management because it generally requires a large amount of energy to maintain indoor thermal comfort. The main objective of this study is to develop a novel method for scheduling AC operations to minimize energy costs and ensure the thermal comfort of occupants under uncertainty. The key challenge is the uncertainty and variability in time-series data and their serial dependence in determining AC operation. To address this challenge, we propose an optimization-informed learning approach that integrates unsupervised and supervised learning techniques with a stochastic optimization model. This method derives energy-efficient and thermal comfort-aware AC operation schedules through a comprehensive interpretation of uncertainties and variabilities in time-series data. Numerical experimental results demonstrate that the proposed approach can reduce energy costs by up to 15.6% and decrease thermal comfort violations by up to 63.6% compared to the Deep Q-learning method, while also reducing energy costs by 1.8% and decreasing thermal comfort violations by 37.5% compared to the forecast data-driven AC scheduling method.
Read full abstract