Application Of Gaussian Process Regression In Meteorological Data

Building energy saving is one of the important links in building low-carbon economy.As a developing country with stable economic growth,the building energy consumption accounts for about 1/3 of the total energy consumption in our country.The simulation and prediction of building energy consumption can guide the implementation of building energy saving measures effectively.The establishment of a reasonable outdoor meteorological parameter calculation model and the acquisition of a typical meteorological year' hourly data for building energy consumption are indispensable links in the simulation process of building energy consumption.The hourly meteorological data in our country is shortage,due to the number of base station is shortage and the location of base station has been moved.Therefore,the development of hourly data and the supplementation of missing data are the focus of typical meteorological year studies.The paper mainly studies the application of Gaussian process regression algorithm in the development of hourly data aim at the four time data and in the supplementation of missing data aim at the short term missing data.Based on the basic theory of Gaussian process regression,the algorithms for the development of hourly data and the supplementation of missing data based on Gaussian process regression are proposed.The effectiveness of the algorithms are verified by applying the two algorithms to develop hourly data and supplement missing data of Baoshan station in Shanghai.Firstly,based on the basic theory of Gaussian process regression,the frameworks of the development of hourly data and the supplementation of missing data are put forward.For temperature and atmospheric pressure,two compound covariance functions are designed to obtain hourly data,and one compound covariance function is designed to supplement the missing data.Meanwhile,the hyper-parameters in the covariance function are determined.The two algorithm is refined by designing the covariance function and determining the super-parameters.Secondly,the above algorithms are applied to develop hourly data and supplement missing data of temperature,atmospheric pressure and relative humidity in Baoshan station,and compare it with the result of cubic spline interpolation method.The experimental results show that: The compound covariance function model is better than the single covariance function model during the supplementation of missing data based on Gaussian process regression.The method to develop hourly data is more suitable for temperature and atmospheric pressure rather than relative humidity.Whether it is a relative parameter or an absolute parameter,the algorithm for supplementing missing data based on the Gaussian process regression is better than the cubic spline interpolation algorithm.Finally,give a 95% confidence interval of the descriptive description to the results.The experimental results show that: the proposed algorithms in this paper improved the accuracy of the hourly data and the supplemental data.It can be used to generate the typical meteorological year hourly data for building energy consumption simulation.