Research On Forecast Model Of Air Conditioning System In Data Center

In the 21 st century,all aspects of China are developing rapidly.The data center as the basis of national information architecture,in order to meet the requirements of Age of Big Data,its number and size are also rapidly growing.The attendance is the increasing in energy consumption in the data center.According to the American Association of Heating and Air Conditioning Engineers(ASHRAE)Technical Committee 9.9(referred to as TC9.9)statistical report shows that the various parts of the electricity distribution in the data center is as follows air conditioning and refrigeration accounted for about 31%,lighting 4%,UPS 8% etc.From the above data can be seen,the electricity consumption of air conditioning refrigeration system accounts the for nearly one-third of the total data center,which is the key indicator to affect energy consumption of the engine room.Therefore,the energy saving optimization of the air conditioning system in data center is a huge potential.At present,China's data center air-conditioning energy-saving technology is mainly reflected in two aspects,on the one hand is the design of energy-saving transformation,so that all aspects can be reasonable operation.On the other hand is the use of automatic control strategy so that the operation status of the air conditioning system can change automatically according to the cold load as well as the outdoor weather conditions.But if only for a single device to control,rather than from the overall consideration,energy consumption problem can not be solved.Therefore,the application of nonlinear predictive control to air conditioning system in data center to achieve energy-saving optimization has great prospects for development.The nonlinear predictive control is based on the control of the forecast model,and the forecast model only emphasizes the function of the model without emphasizing the specific structure of the model.Because the air conditioning system has the difficulty of establishing mechanism modeling,so it is very important to study the forecast model of air conditioning system.In this paper,the forecast model of the air conditioning system in data center in Beijing is studied in the summer.First,the characteristics of the air conditioning system in data center are summarized.Combined with the characteristics of the air conditioning system in data center and the current research to the forecast model,the basic algorithm to establish the forecast model of the air conditioning system in data center is proposed-support vector regression machine.The support vector machine algorithm based on statistical theory and the principle of the algorithm are introduced in detail,and the foundation stone is established for the forecast model.The main contents of the study include the following aspects:(1)This paper learns the form,characteristics and operating conditions of the air conditioning system in data center.The air conditioning system in data center deeply are studied deeply to determine the input and output parameters of the forecast model of the air conditioning system in data center.Through learning and analysis,determine the algorithm to establish the forecast model of the air conditioning system in data center.(2)The establishment of the forecast model requires a lot of data.In this paper,the data sampling strategy is established according to the control requirements and the operatingcharacteristics of the air conditioning system.The model of the collected data is established in TRNSYS,and the automatic operation acquisition is realized and the data is preprocessed.(3)The support vector machine(SVM)is used to identify the forecast model of the air conditioning system in data center.The contents conclude the selection of the kernel function,parameter optimization,support vector regression model selection and other work content.At the same time,the forecast model based on BP neural network is simulated and compared with the forecast model based on SVR in MATLAB.(4)The off-line forecast model is online corrected in MATLAB.In this paper,according to the characteristics of the air conditioning system in data center determines the training sample logarithm,training time and the specific online correction algorithm.