A comparative analysis of Deep Neural Networks and Gradient Boosting Algorithms in long-term wind power forecasting

Abstract

A vital step toward a sustainable future is the power grid's incorporation of renewable energy sources. Wind energy is significant because of its broad availability and minimal environmental impact. The paper presents a comparative analysis of recurrent neural network algorithms and gradient boosting machines applied to time series data for the regression issue of estimating the active power generated by a wind farm. Gradient boosting algorithms combine the advantages of a few machine learning models (decision trees, random forests, etc.) to produce a powerful prediction model. In addition to conventional recurrent neural networks, the article deals with long short-term memory and gated recurrent unit as cutting-edge models for time series analysis and predictions. A comprehensive analysis was carried out on a large wind power generation data set.

Keywords

• machine learning
• recurrent neural network
• long short-term memory
• gated recurrent unit
• gradient boosting machines
• xgboost
• wind farm
• power generation