Research On Control Methods Of HVAC Systems For Smart Commercial Buildings

With the rapid development of urbanization,the share of energy consumption of commercial buildings(e.g.,offices,markets,restaurants)in total building energy consumption is increasing.In commercial buildings,energy consumption of HVAC systems account for about 45% of total energy consumption,which brings a great economic burden to commercial building operators.In order to conserve energy consumption/energy cost,the most direct way is to reduce the input power of the HVAC system,but it may affect thermal comfort of occupants.Therefore,it is very necessary to reduce energy consumption/ energy cost on the premise of ensuring thermal comfort for occupants.However,the challenges to achieve above goal are as follows: uncertain system parameters,unknown explicit building thermal dynamic model,temporally and spatially coupled constraints related to indoor temperature/carbon dioxide concentration,and a large discrete solution space.To overcome the above challenges,this thesis studies the control methods of HVAC systems in commercial buildings.Firstly,this thesis investigates an energy cost minimization problem of HVAC systems in multi-zone commercial buildings.Then,we propose a novel control method of HVAC systems based on multi-agent deep reinforcement learning for solving this problem.Note that the proposed method does not require any prior knowledge of uncertain parameters and explicit building thermal dynamic models.Extensive simulation results based on the real-world traces show the effectiveness,robustness and scalability of the proposed method.Compared with the rule-based scheme and heuristic scheme,the proposed method maintains thermal comfort for occupants while reducing energy cost by 75.25% and 56.50%,respectively.Moreover,this thesis investigates an energy cost minimization problem of HVAC systems in multiple time-scales.Specifically,in view of the frequent changes of valve position and supply air temperature will have degradation effects on the lifetime of equipment,valve position and supply air temperature are scheduled on a large time scale,while the regional air supply rate is dispatched on a small time scale.In order to solve this problem,we propose an HVAC system control method based on two-timescale deep reinforcement learning.Extensive simulation results based on the real-world traces show the effectiveness and robustness of the proposed method.Finally,this thesis makes a summary and points out future research directions.