Modelling spatially and temporally correlated wind speed time series over a large geographical area using VARMA

Artikel i vetenskaplig tidskrift, 2017

This study presents a modified vector auto-regressive moving average (VARMA) modelling procedure to model spatially and temporally correlated wind speed time series over wide geographical areas. The standard VARMA is normally used to model stationary time series with Gaussian distribution. However, wind speed is non-stationary (mean and variance varies over time) and non-Gaussian. Hence, a method that can be used to transform wind speed data into a stationary and Gaussian time series is introduced in the modified procedure. To show the applicability of the procedure for different scenarios, six cases are investigated in the North and the Baltic Sea. The results show that the procedure can be used to model spatially and temporally correlated wind speed over a large geographical area. In addition, the resulting model can capture probability distribution and periodic characteristics of the wind speed data. Furthermore, based on the investigated case, it is shown that a vector auto-regressive model of order three is a reasonable model structure which can be used to model spatially and temporally correlated wind speed in the North and the Baltic Sea area provided that the power transformed wind speed data is normalised by its monthly mean value and its variance.