Wind Speed Prediction using Machine Learning Techniques

Abstract

Wind energy is a free source of renewable energy. In order to ensure that the wind energy is utilized to its maximum extent, the wind energy potential must be evaluated. For this, we need to forecast the wind speed as wind energy is directly proportional to the cube of wind speed. But wind speed forecasting is very difficult as there are issues like scheduling of power system along with control of wind turbine dynamically. To this aim, we investigated different Machine Learning algorithms, namely Autoregressive Integrated Moving Average (ARIMA), Long Short-Term memory (LSTM) that helped in sequential data prediction, linear tree regression, decision tree and random forest. The data was recorded from a place where a device was located in an empty area, at 21M. The results were evaluated on the basis of three metrics namely, Mean Absolute Error, Mean Square Error and Root Mean Square Error. The result shows that LSTM has the least Mean Absolute Error but there is a scope of improvement in terms of accuracy that are obtained in comparison to the persistent method. Moreover, the selection of the best algorithm in order to have the best result also depends upon the data sources. This study helps in accurate wind speed forecasting that becomes a necessity to schedule dispatchable generation in the day-ahead electricity market.