Research And Application Of Latent Heat Flux Data Gap Filling Based On Extreme Learning Machine

Latent heat flux is an important part of the Earth’s surface energy budget.It’also play an important part of the hydrological cycle.The complete latent heat flux data is an important parameter for estimating crop growth models and hydrological models.For accurately calculating crop water requirements,agricultural water regulation and regional water resources management.It also has a important meaning.However,due to weather and other reasons,the eddy correlation meter for detecting latent heat flux data will be damaged to varying degrees,resulting in the lack of data,and it is difficult to provide complete latent heat flux data.Therefore,latent heat flux data gap filling has become an urgent problem to be solved.Traditional latent heat flux gap filling models require a large amount of meteorological data,And the calculation process is complicated.Gap filling will become tricky when data is scarce.The purpose of this paper is to investigate the use of machine learning methods to interpolate the missing data of latent heat flux in the absence of meteorological data.The main research contents of this thesis include:(1)Using the machine learning method to analyze and select the latent heat flux influence factors,analyze and demonstrate the results from the ecology.(2)Based on the selected impact factors,using the extreme learning machine,BP neural network,support vector regression and XGBoost to interpolate the missing data,According to the result,ELM and XGBoost are the most effective,R2 reached 0.8672 and 0.8504 respectively,but the ELM is more convenient in parameter adjustment.Therefore,the ELM is selected as a further research method.(3)The initial parameters of the extreme learning machine are random leads to unstable results,when there are too many hidden layer nodes,the over-fitting situation will occur.In order to improve the generalization performance of the ELM,the kernel extreme learning machine is introduced,and the kernel extreme learning machine is used after the parameter tuning using the genetic algorithm.The R2 reaches 0.8545.Based on the characteristics of latent heat flux,the time-series experiment is carried out.The result shows that the best R2 can reach 0.8836,which is a more suitable latent heat flux gap filling strategy.