| The purpose of air conditioning is to provide the expected temperature field for indoor occupants,nowadays,return air temperature based control is still the mainstream HVAC control method.According to the Lagrangian description of flowing process,return air temperature represents the average temperature of upstream converging air,not the actual temperature perceived by the users.Air flow process is limited by higher-order nonlinear governing equations,using full-scale computational fluid dynamics simulations to predict temperature distribution will exhibit excessive consumption of computational resources and severe hysteresis.Accurate,fast and low-cost access to obtain indoor temperature distribution is an important issue in the field of HVAC control.Researchers tend to recall CFD models with either steadily or pseudo-dynamically manners,avoiding frequent numerical iterations of Navier-Stokes equations.However,the dilemma of high computational cost and slow calculation process for temperature distribution prediction has not been solved.The proper orthogonal decomposition method can quickly achieve reduced-order representation of flow process,the limited modal after interception takes up only a tiny amount of memory space.Through the mining of the temperature distribution data inside the room,the reduced order model learns to obtain high-dimensional mechanism between supply air temperature and the thermal condition at specified location in the room which could shorten the time of temperature prediction.Firstly,we create a three-dimensional model for the office room and make some reasonable simplifying assumptions about the flow process,and then set boundary conditions at various locations in the space.The CFD software is used to calculate the unstructured grid of the room object,and the distribution of the room temperature field under different supply air temperature settings is obtained by numerical simulation.In addition,the root cause of the hysteresis of computational fluid dynamics simulation is pointed out by analyzing the mathematical properties of the flow control equation.16,326 sets of temperature information will be used as training data for the POD model.Secondly,the modeling mechanism of POD model is given,and a reduced-order network composed of observation layer,deviation layer,modal layer and modal coefficient layer is constructed.In this way,POD model is established for the office room.The first 6 major modals intercepted according to energy ratio criterion will be used for temperature field reconstruction and generalization.It is calculated that the generalized accuracy of the POD model is 89.1% in temperature prediction,which shows its feasibility for temperature prediction.Thirdly,the source of the prediction error caused by vector interpolation method to the modal coefficients is analyzed.Ensemble learning model consisting of weak learners was used for modal coefficient prediction and top 10 significant related features obtained by Pearson correlation analysis was introduced in the ensemble learning framework.The generalization results show that the ensemble learning model helps to increase the accuracy of modal coefficient prediction by 8.9%.Finally,considering the non-full occupants scenario that occurs during the actual use of the office room,an adaptive air supply control logic based on office pattern recognition is proposed.The POD model is used as virtual sensing layer in the control logic,and the temperature of supply air is used to predict the temperature distribution on the observation nodes.Then,genetic algorithm in the decision layer will solve the defined temperature distribution objective function and send the optimized air supply parameters.The results show that the adaptive control logic will show significant advantages compared to the return air temperature control in the case of uneven distribution of office occupants. |
PDF Full Text Request