Modeling and Forecasting of Wind Speed Time Series in the United Arab Emirates

  • Hye Yeon Kim

Student thesis: Master's Thesis

Abstract

As in many other parts of the world, the demand for renewable energy has continued to grow in the United Arab Emirates (UAE) with increasing local demand for energy. In particular, there has been a growing interest in wind resources. Evaluating wind energy potential understanding trends and forecasting changes in wind are the key tasks in efficient implementation of wind energy projects. In this study, time series approach is adopted to model and forecast the wind speed, for both short term and long term forecasting. The time series data used in this study was constructed by taking a monthly average of observed 10-minute wind speed time series in the United Arab Emirates (UAE), measured from the international airport of Abu Dhabi at 27m height, from year 1982 to 2010. Due to the apparent seasonality shown in the wind speed data, periodic models were selected. The time series models used include autoregressive (AR), autoregressive moving average (ARMA), and periodic autoregressive integrated moving average (PARIMA). The AR component was employed because of the assumption that wind speed is correlated to the previous values. The MA component was added because it helps consider previous forecasting errors in making prediction. The integration component was applied by subtracting lagged time series from the original, to make the data more stationary. All the models were applied to each monthly time series, to study different behaviors of wind in each month, and different combinations of parameters were tested to obtain optimal models. The forecasting performance of all the models is evaluated with the root mean square error (RMSE), relative root mean square error (rRMSE), bias, relative bias values of each model. Results indicate that the PARIMA models provide a good fit to the wind data in Abu Dhabi. Then residual modeling with generalized autoregressive conditional heteroskedasticity or GARCH(1,1) distributed variance of the residuals was attempted to improve the forecasting results since it was shown that residuals from previous modeling still includes some information about the data. It was concluded that GARCH model could be useful in modeling changing fluctuations around the variance of wind speed.
Date of AwardMay 2015
Original languageAmerican English
SupervisorPrashanth Marpu (Supervisor)

Keywords

  • Renewable Energy
  • Wind Energy
  • Forecast The Wind Speed
  • Long Term Forecasting
  • Periodic Autoregressive Integrated Moving Average.

Cite this

'