Influence of Deep Learning on Precision Improvement in Predictive Models of Wind Power Generation

This paper, proposes the use of Deep Learning in predictive nonparametric models that use artificial intelligence tools to approximate power curves of wind farms. Three different tools are evaluated: artificial neural networks, fuzzy inference systems and Auto Encoders, an initial model of deep learning networks. The tools are inserted in a non-parametric model of power prediction, where they are compared. The results show that the autoencoder-based power curve performs well above other proposed tools. This significantly improves the performance of the predictive power model.

I. Introduction

Due to the growth of the energy demand in Brazil, the wind power source has been a very plausible solution, but this one has a basic characteristic, the variability of wind, which generates the possibility of new heuristics focused on the wind power forecast. Investigating new possibilities, the predictive power models, can be linked to parks or wind turbines, whose type can be parametric or non-parametric. In the case of non-parametric models, the highlight go to the predictive models that tend to adjust the power curve. For the adjustment of power curve in the predictive model, artificial intelligence tools are used, as: Artificial neural networks, Fuzzy inference systems, and new study tendencies such as the Deep Learning models. These tools have an essential role to aggregate accuracy of the response of predictive models.