New Deep Learning-Based Approach for Wind Turbine Output Power Modeling and Forecasting

Abstract

An intelligent machine learning-based method is developed in this paper for modeling and prediction of the wind turbine (WT) output power. The developed technique makes use of the advanced machine learning models for creating prediction intervals around the WT power samples. Compare to the other techniques which make a point-by-point prediction, the proposed technique can capture the uncertainty of the forecast error using generating lower and upper bounds around the samples. Moreover, a fuzzy min-max technique is employed to get into a Nash equilibrium between the prediction confidence level and the average prediction bandwidth. In order to get into reliable results, two support vector machine is deployed which compete with each other to improve the prediction performance. This game can potentially contribute to a highly accurate prediction model. Also, a new optimization method based on the flower pollination algorithm (FPA) is developed due to the high complexity and nonlinearity of the modeling, which can search for finding the most optimal structure of the model. The feasibility and appropriate performance of the model is assessed using the experimental data from Australia islands. IEEE