Development of wind energy prediction models using statistical, machine learning and hybrid techniques: a case study

This paper presents the development of statistical, machine learning, and hybrid models to predict the wind energy generation of a major wind farm in Sri Lanka named Nala Danavi. Regression based statistical techniques, namely, Multiple Linear Regression and Power Regression were applied to the input variables of wind speed and ambient temperature producing expressions for wind energy in terms of those weather indices. Similarly, the machine learning techniques of Support Vector Regression and Artificial Neural Network were applied in developing another set of wind energy prediction models. Unlike in the methods mentioned above, the Support Vector Machine was applied only to the past energy data. In contrast, a Vector Error Correction model was developed by using both wind energy and weather data. Further, time series modelling of wind energy data was performed to develop a Seasonal Autoregressive Integrated Moving Average model. The research was further extended to develop two hybrid prediction models by combining Support Vector Machine and Artificial Neural Network each with Seasonal Autoregressive Integrated Moving Average. The performance of all the models was assessed and compared in terms of the Coefficient of Determination, Root Mean Square Error, and the Mean Absolute Error. As per the results, all the models were highly accurate while the Support Vector Regression produced the most precise prediction model. The models developed in this research can be used to predict the wind energy generated at Nala Danavi wind farm using either the previous energy data or the projected weather data of the region.