Machine Learning for the Power Generation Forecast of a Wind Farm

Abstract

One of the greatest challenges of the wind energy nowadays is the delivery of its power output into the energy grid, because of the intermittency of the wind speed and the fluctuating nature. For that reason, an accurate forecast for the short-term period is necessary to increase the insertion of wind power into the energy mix, as well as preventing extreme events and other possible drawbacks. In this regard, Machine Learning algorithms have played an important role in the wind power prediction in recent years, since this automatic learning method presents several advantages that make it ideal for this task. In this study two Machine Learning approaches will be studied and developed with Python, the Linear Regression algorithm and the Multilayer Perceptron algorithm, which is a kind of Artificial Neural Network, applying them to the dataset with real measures (wind speed, power generation, temperature,…) of an actual wind farm for a two-year period as a case study. The two algorithms will have multiple variables of the set as inputs in order to learn from the existing data, train the corresponding algorithm, so it can be utilised to forecast future wind power generation. Both models will be validated with the aim of verifying the accuracy of the methods.