A novel stochastic model for very short-term wind speed forecasting in the determination of wind energy potential of a region: A case study from Turkey

Abstract

The dominant use of fossil fuels in energy production and the great damage they have caused to both environment and human health are nowadays well-noticed. This dominant use of fossil fuels brings with it problems such as energy shortage, environmental pollution, and ecological degradation. Recently, there has been a tremendous turn towards renewable energy sources in the future energy investments of the governments. Of renewable energy sources, wind energy has located at a significant owing to its abundance, infinity, independence, reliability, and clearance characteristics. However, it is of great importance of forecasting both the amount and prices of electricity for suppliers, particularly in intraday markets, because it is a serious problem for suppliers to generate more or less electricity than they guarantee. They may face sanctions or lower profit margin risks owing to unbalances between the forecast and generate values. Accordingly, a novel model - variance sensitive exponential smoothing model - (VSES) was proposed in the present study and then compared with three well-known models in the literature namely Trigg & Leach (T-L), Pantazopoulos and Pappis (P-P) and optimized simple exponential smoothing (o-SES). Then the wind speed dataset of Çeşme district (in İzmir, Turkey) in 30-minute, 1-hour, and 3-hour time horizons are tested with these four models. In the comparison of the models, four statistical benchmarks (MAD, MSE, MAPE, and RMSE) are discussed. In the results, the RMSE value of all models is smaller than 0.9 m/s for all forecasting horizons. VSES model proposed in this study has exhibited the very satisfying MAD, MSE, MAPE, and RMSE values of 0.318 m/s, 0.183 m/s, 14.60%, and 0.427 m/s, respectively for 30-minute and 0.382 m/s, 0.258 m/s, 16.29%, and 0.506 m/s, respectively for 1-hour, respectively. On the other hand, the best model for 3-hour time horizon is achieved for the P-P model with the best MAD, MSE, MAPE, and RMSE values of 0.574 m/s, 0.543 m/s, 26.38%, and 0.729 m/s, respectively, and it is closely followed by the VSES model with 0.601 m/s, 0.614 m/s, 26.72%, and 0.775 m/s, respectively. In conclusion, it is clearly observed that the VSES model proposed in this study can be successfully applied for the determination of the wind energy potential of the regions, and effectively used in intraday markets particularly for very short-term wind speed forecasting.