Leveraging provision of frequency regulation services from wind generation by improving day-ahead predictions using LSTM neural networks

The growing penetration of volatile renewable generation is substantially impacting the operation of power systems, which results in increased balancing needs. Since wind turbines are equipped with power-electronic converters, they can be used to provide such frequency regulation reserves. However, in a competitive environment in which reservation of ancillary services capacity has to be carried in day-ahead, their provision is hampered by the limited knowledge on the energy that will be available with sufficient reliability. In this context, this paper aims at estimating the impact of prediction accuracy on the ability of wind turbines to increase their profitability by delivering power reserves. To that end, an advanced recurrent neural network architecture known as Long Short Term Memory is used to provide high-quality predictions. Simulations demonstrate that forecast accuracy is a key element to increase the economic value of wind farms, ultimately fostering their large-scale deployment by lowering integration costs.