A Two-Stage Feature Extraction Gated Recurrent Unit for Wind Power Prediction

With the rise of environmental awareness, approaches of power generation other than traditional ones are receiving much attention. Therefore, renewable energy power generation and its integration with other power generators in a smart grid is becoming an important issue. In an environment with the nature of dynamic and real-time, data increases both in volume and variety. There are studies in literature using deep learning models to make data prediction, and many of them have good results. However, issues such as overcomplicated data dimension may occur when deep learning models are dealing with big data. This study proposes a model combining association analysis, sequence analysis, and Gated Recurrent Unit (GRU) to predict the amount of power generated by wind turbines in Taiwan. Our experiment shows that association rule analysis keeps computation efficiency and also improves model accuracy at the same time.