Posters

Abstract

Wind energy conversion systems such as wind farms and power plants need to be managed for regular scheduling of grid load and power generation balance. Wind energy being a dynamic resource, as any other renewable, has intermittent nature. In the present study, a detailed and comprehensive analysis is put forth to discuss the application of Artificial Neural Networks in 10-minutes resolution wind speed forecasting. Four different sites in Pakistan, based on promising wind resource potential are targeted, namely Sanghar, Sujawal, Tando Ghulam Ali and Umerkot. One-year ESMAP data is used for wind speed estimation at 80m altitude, with temperature, pressure, air density, relative humidity and wind speed at other altitudes used as input. The statistical performance metrics used include Mean Absolute Error, Mean Squared Error and Pearson Correlation Coefficient. The study found that Sanghar performed the best with R-value of 0.9854 and Mean Squared Error of 0.2621 m/s. Umerkot performed the worst with 0.9798 R-value and Mean Squared Error of 0.3129 m/s. The results thus obtained show that each site typically requires tuning of the Artificial Neural Network hyperparameters and that one model may not simply be deployed at another site. The results were compared with Windographer results obtained from data reconstruction and analysis. Artificial Neural Network outperformed the latter with an average improvement of 2.436% in R-value. This indicates the importance and practical robustness of using Artificial Neural Networks for power plants operations.