Estimating the short-term and long-term wind speeds: implementing hybrid models through coupling machine learning and linear time series models

Abstract

Wind speed data are of particular importance in the design and management of wind power projects. In the current study, three types of linear time series models including autoregressive (AR), moving average (MA), and autoregressive moving average (ARMA) were employed to estimate short-term (i.e., daily) and long-term (i.e., monthly) wind speeds. The required data were gathered, respectively, from the Tabriz and Zahedan stations in the northwest and southeast of Iran. The MA models outperformed the AR and ARMA on the both daily and monthly scales. Daily and monthly wind speed values, as a function of lagged wind speed data, were then estimated using two machine learning models of random forests (RF) and multivariate adaptive regression splines (MARS). It was found that the RF and MARS provided similar results; however, RF performed slightly better than the MARS. Finally, the stand-alone time series and machine learning models were coupled to improve the accuracy of the wind speed estimation. Accordingly, the hybrid RF-AR, RF-MA, RF-ARMA, MARS-AR, MARS-MA, and MARS-ARMA models were implemented. It was concluded that, the hybrid models outperformed the stand-alone RF and MARS for both short- and long-term wind speed estimations where, the RF-AR and MARS-AR hybrid models provided the best performances. The hybrid models tested in the present study could be effective alternatives to the stand-alone machine learning-based RF and MARS models for the estimation of wind speed time series.