Short Term Wind Speed Forecasting : A Machine Learning Based Predictive Analytics
The challenges posed by the intermittence and uncertainty of renewable energy due to its variability and limited storage require accurate forecasts for economies looking to source a significant amount of energy from renewables. We report on the use of several supervised learning models such as Random Forest, Extremely Randomized Trees, Support Vector Regression and k-Nearest Neighbors Regression to forecast ahead of time wind speed measurements using data from the wind met masts located at Buguey, Ballesteros and Sta. Ana, Cagayan. Results show that in terms of predicting the next hour wind speed measurements for one day, the k-NNR model outperforms the other three models while the ET model have shown the highest predictive performance among the four models in prediction of the next hour wind speed measurements for one month and 20% of the total data. It is anticipated that the proposed ET model can be used as an effective wind speed prediction model as well as the k-NNR model. The common perception by energy companies in ASEAN that RE output is unpredictable needs to be rethought in the sight of the new AI techniques.
A. J. Domingo, F. Carlo Garcia, M. L. Salvaña, N. J. C. Libatique and G. L. Tangonan, "Short Term Wind Speed Forecasting : A Machine Learning Based Predictive Analytics," TENCON 2018 - 2018 IEEE Region 10 Conference, Jeju, Korea (South), 2018, pp. 1948-1953. doi: 10.1109/TENCON.2018.8650287