Heteroscedastic Spatio--Temporal Model And Its Application In Wind Speed Prediction

In recent years,wind power generation has great value and application prospect in the fields of new energy and green energy.The wind speed data is non-negative,biased,heavy-tailed and containing many zero,which makes us face great challenge in short term wind speed prediction and model construction.Based on the above anal-ysis,we propose an improved model to first-order lag multiple linear model in this thesis.Firstly,in the spatial dimension,we use the meteorological stations in the same area as the space covariates to perform a first-order lag multiple linear spatio-temporal fitting.In the above spatio-temporal fitting error term,given an interpretable atmo-spheric covariate,we propose a heteroscedastic spatio-temporal prediction model based on the t distribution.Secondly,in the process of statistical inference of parameters,we propose a two-step inference algorithm.In the first step,we use the Smoothly Clipped Absolute Deviation(SCAD)penalty function to make variable selection and parameter estimation of spatial covariates in the spatial dimension.In the second step,we use the Expectation Maximization(EM)algorithm for the likelihood function of multivari-ate t distribution.Finally,we apply the proposed model in this thesis to public data sets,and the results indicate that there are significant improvements in forecasting accuracy.