Simulation Study On Air Conditioning Temperature Prediction Based On Long Short-term Memory Neural Networ

With the growth of energy demand and the development of artificial intelligence technology,combining artificial intelligence to predict and control energy management systems has become a hot topic.Global energy statistics show that buildings consume about 40 per cent of total energy,HVAC systems consume more than 50 per cent of commercial building energy and have considerable potential for energy conservation.Data processing through deep learning can assist HVAC systems to achieve multiobjective control,including reducing energy consumption,improving system efficiency,ensuring user thermal comfort,cutting peaks and filling valleys,etc.The simulation of air conditioning model and prediction model can analyze the energy-saving potential and cost of control strategy and algorithm before applying the new strategy.The main content of this study is to predict the energy consumption,indoor temperature and user thermal comfort of air conditioning system by using the Long Short-Term Memory(LSTM)neural network.The effect of LSTM predictive control on the operation of air conditioning is simulated.(1)Taking Qingdao as a model and combining the design specifications of heating,ventilation and air conditioning for civil buildings,a year-round load model of building and air conditioning system was established by using Sketch Up and Open Studio.Models were imported into Energy Plus with corresponding data interfaces for simulation studies based on local climate conditions in Qingdao provided by Meteonorm and WMO,and empirical formula models and Simscape air conditioning models were established in Matlab / Simulink for comparison.(2)The air conditioning operation data were divided into normal and random start and stop states.Each data was scaled again into training sets,test sets,and validation sets for BP neural network,support vector machine,and LSTM neural network training and prediction,respectively.The influence of multi-layer LSTM neural network combining convolution neural network and attention mechanism on prediction accuracy was also tested,and LSTM neural network was significantly more accurate than several other prediction methods.(3)Establishment of Matlab and Energy Plus simulation models to update neural network status,adjust control strategies and control objectives in real-time control.A two-layer parallel LSTM neural network model is proposed to divide the prediction error into two parts,improve the transparency and stability of the model,reduce the intrusion dependence of data and improve the adaptability of the model.(4)Compare the predictive control of LSTM neural network with the air conditioning temperature control of PID to compare the control effect of user thermal comfort.Various temperature and thermal comfort indexes and corresponding neural networks are presented to compare the variation of indoor temperature under different air conditioning control modes.Based on the measured data in the literature,the difference between the actual data and the simulated data is compared.