Study On The Energy Efficiency Prediction Model Of The Plant Of Central Air-conditioning System

As the COP of the air conditioning system is an important indication of system performance, this paper uses modeling methods to predict COP in order to give advices for the Fault diagnosis and to improve the system efficiency, and on the whole to save the energy.In order to get the operational data of the chiller in a building during the air-conditioning season, this paper using the energy consumption analysis software EnergyPlus to simulate the air condition system. Then using the BP neural network to build the COP prediction modle of the unit, and finally verify the reliability of the modeling method.As the aim of this paper is to predict the COP of the uint, the output of the model is the COP at the t+1 memont. By using the method of regression analysis, the four parameters that most significantcorrelation to the COP are the unit power, the chilled water inlet and outlet temperatures, and the cooling water outlet temperature at t moment. Finally by determining the inputs of the prediction model are the parameters mentioned above both at t and t-1 moment, the basic structure of the model is finished.This paper uses three kinds of methods to improve the precision of the model. First is to increase the adaptive learning rate to overcome the shortcomings of the BP network. For example, the system is unstable as training and some time the results converge to a local minimum. Second is to adjust the following 3 parameters that have the biggest influence on the prediction ability of BP network model, the number of input layer nodes and hidden nodes, the mean square error. Third is to change the input mothod so that the model can learn the operation rule of the whole air conditioning operation period. By using the methods mentioned above, the prediction accuracy is greatly improved. The absolute errors of expectations and network outputs are mostly concentrated in -0.1 to 0.1 interval, while the relative errors are mostly concentrated in -0.05 to 0.05 interval.To verify the reliability of the model methods mentioned above, this paper use following three different systems operation data to build three models: the model provided in the EnergyPlus, the model created by the Matlab, and the actual air conditioning system. The results shows that this modeling method is reliable.