Application of Machine Learning on Wind Blowing Speed

Abstract

The estimation of the meteorological parameters such as temperature, pressure, humidity, sea wave height, sea wave speed, and wind speed gains importance as it provides us with information about the unknown future values of these parameters. In this study, the measured sea shore wind speed data is predicted using a variety of machine learning algorithms. Accordingly, these wind speed data are obtained considering daily measured wind speed values of a sea shore region found in Turkey. A total of 1,415 data found in the wind speed data cluster has been utilized in forecasts depending on the historical-time series. The algorithms used for this purpose include long-short term memory, adaptive neuro-fuzzy inference system with fuzzy c-means (FCM), subtractive clustering (SC), and grid partitioning (GP). A cumulative of 66 different models have been structured using these four tools. The quality of the forecasts has been compared according to the statistical accuracy errors comprising of mean absolute error (MAE), root mean square error (RMSE), as well as the correlation coefficient (R). In this context, the analyses of the statistical errors of the computations have indicated that the best wind speed forecast for this sea shore region is obtained under the LSTM tool. Ultimately, the MAE, RMSE and R values of this best model were calculated to respectively correspond 0.2477 m/s, 0.3315 m/s, 0.9951.