| Building energy consumption accounts for around 30-40% of a country’s total energy consumption.In building energy consumption,HVAC systems account for around 40% of energy consumption.Although the HVAC system consumes a high amount of energy,the achieved thermal comfort for occupants is typically low.The reason for the above phenomenon is that HVAC control algorithms can not accurately know the ture thermal comfort of occupants.Consequently,effective decisions can not be made to ensure high occupant thermal comfort and low system energy consumption.Traditional thermal comfort models(e.g.,PMV)are used to predict the level of thermal comfort in groups,which are less accurate when applied to individuals.With the development of Io T sensing and machine learning technologies,new opportunities for user thermal comfort modeling have arisen.Although some progress has been made in existing research,the following shortcomings remain:(1)Large samples are required to train a personal thermal comfort model;(2)Most of thermal comfort control methods do not take into account individual thermal comfort models.Therefore,it is necessary to study the methods of personal thermal comfort modeling and control methods based on machine learning.To reduce the dependency of personal thermal comfort models on a large number of training samples,this paper presents a personal thermal comfort modeling method based on deep metalearning.The algorithm highlights the similarity between the known thermal comfort samples and the samples to be predicted by introducing a cross-attention module in the deep neural network.Since deep meta-learning supports few shot learning,small amount of thermal comfort feedback data is required.Therefore,it has a wider applicability.Simulation experiments show that the proposed method improves the prediction accuracy by 22.86%-42.86% compared to other baselines under the same data.To overcome the drawbacks that most existing studies on HVAC control did not consider the use of individual thermal comfort models,this paper incorporates the developed individual thermal comfort model into the closed-loop control of HVAC systems,and formulates an HVAC system operation cost minimization problem.Due to the existence of model uncertainty,parameter uncertainty,and temporally-coupled operational constraints,the above optimization problem is reformulated as a Markovian decision process.Then,a personal thermal comfort control method based on Deep Deterministic Policy Gradient is proposed.The method does not require the prior knowledge of uncertain parameters and an explicit model of building thermal dynamics.Simulation results show that the proposed control method can reduce operating cost by 4.69%-16.45% compared to other baselines at the premise of maintaining high thermal comfort. |
