Research On Decoupling Prediction And Optimal Control Of Heat And Humidity Load In Dedicated Outdoor Air System

As living standards continue to improve,the demand for high-quality indoor environments has been increasing.Building air conditioning,as the primary equipment for creating a comfortable indoor environment,is also accompanied by an increase in energy consumption.To address this issue,dedicated outdoor air system(DOAS)have been developed,which not only meets indoor air quality requirements but also achieves the goal of building energy conservation.However,DOAS operates differently from conventional air conditioning systems,and the existing total load forecasting control methods for air conditioning systems do not fully meet the requirements of efficient DOAS operation.The decoupling prediction of heat and humidity load of the air conditioning system has been proposed as an effective means to realize energy-saving DOAS operation.(1)This paper explores the decoupling prediction of DOAS heat and humidity loads.Specifically,it analyzes the thermal and humidity causes and load composition of DOAS,and builds a DOAS radiation cooling laboratory in Guangzhou as the research object.The Energy Plus DOAS model is used to simulate the DOAS operating load data in the cooling season in Guangzhou,which provides the data basis and input characteristics for the decoupling prediction of heat and humidity load.(2)The long and short term memory neural network(LSTM)model,multiple linear regression(MLR),and multi-layer perceptron(MLP)are built,and the models are trained based on air conditioning load data from Guangzhou’s cooling season.Heat and humidity load decoupling prediction is carried out,and the performance of the three load decoupling prediction models is compared and analyzed by evaluation indices.Results indicate that the LSTM decoupling prediction model has the best performance,with mean absolute percentage error(MAPE)9.1% and 4.8% lower than MLR and MLP,respectively.Root-mean-square error(RMSE)was also found to be 225 W and 141 W lower than MLR and MLP,respectively.(3)Combining the advantages of different neural network structures,the attention mechanism(AM)and convolutional neural network(CNN)are combined with LSTM to create the proposed hybrid model,AM CNN LSTM(ACL).The ACL model enhances the mining ability of hidden characteristics of load history data and increases the weight allocation ability of input characteristics.The air conditioning load data from Guangzhou’s cooling season is used for training and forecasting.Compared with LSTM,MAPE decreases by 1.4%,and RMSE decreases by 60 W.Comparative error analysis shows that ACL is more stable than LSTM,and the maximum error is reduced by 14%,proving the advantage of the ACL decoupling prediction hybrid model.(4)A co-simulation platform of Energy Plus and Python is built,and the ACL load decoupling prediction model is used to achieve online monitoring and control.Energy-saving analysis based on ACL load decoupling predictive control is conducted using summer design days in Guangzhou as an example.The results demonstrate that the proposed method can reduce power consumption by 7.7%.