Research On Gaussian Process Regression Model Applied In Kunming Monthly Average Temperature

Weather changes have a huge impact on agriculture and people's going out,so people have paid great attention to weather prediction since ancient times.In recent years,the trend of global warming has intensified,extreme weather around the world has frequently occurred,and all kinds of bad results brought about by rapid changes in temperature have warned us that we must put forward stricter standards for temperature forecasting.However,because the fluctuation of temperature has uncertain random characteristics,and the trends and seasonal characteristics may not be obvious,the prediction effect of traditional time series models have become limited.Therefore,it is imminent to use novel algorithms to make more accurate temperature predictions.In this paper,a Gaussian process regression algorithm is applied to the prediction of the monthly average temperature in Kunming.Considering the complexity of monthly average temperature fluctuations,this paper proposes a combined kernel function based on the characteristics of the monthly average temperature in Kunming,and based on the combined kernel function a forecast for the 21-month average monthly temperature in Kunming was made.The main work of this article is:(1)The monthly average temperature of 221 months in Kunming from January 2001 to May 2019 was data-analyzed.It was found that the temperature fluctuations in Kunming in the past 19 years were not large,and no year-on-year increase in the temperature in Kunming was found.(2)Starting from the weight space method and the function space method,the prediction mean and prediction variance of the Gaussian process regression are derived,and the intermediate derivation process is explained in detail and necessary proofs.(3)Considering the limitations of traditional time series models,a new algorithm of Gaussian process regression is introduced into the prediction of monthly average temperature.Compared with traditional autoregressive moving average models and decision tree regression,Gaussian process regression has higher prediction accuracy and can give confidence intervals for predictions.(4)The choice of kernel function in Gaussian process is discussed in detail in this paper.The various kernel functions and their properties are introduced,and a sample image of the kernel functions is given.Then,the influence of parameter transformation on the function structure of kernel function and the influence of the kernel function parameters on the predicted distribution are analyzed by digital simulation.And the properties of single kernel function and combined kernel function and their influence on the predicted distribution are studied respectively,which can be used as a reference for the subsequent modeling theory.(5)Considering the characteristics of complicated monthly average temperature change in Kunming,a combined kernel function based on the characteristics of this data set is proposed,and the combined kernel function is used to predict the monthly average temperature of Kunming.Experimental results show that the Gaussian process regression algorithm based on the combined kernel function has better prediction accuracy than the classical autoregressive moving average model.