Wind Power Prediction considering Ramping Events Based on Generative Adversarial Network

In view of the growing depletion of traditional fossil fuels and their adverse impact on natural environment, wind energy has gained increasing popularity across the globe. Characterized by wide distribution, low cost, and well-rounded technology, it has achieved fast-growing installed capacity in recent years. However, wind power is volatile and random in nature and the power ramping events caused by extreme weather always threaten the safe, stable, and economic operation of the power grid. To address the problems of insufficient sample data and low prediction accuracy in existing ramping prediction methods, a new way of wind power prediction considering ramping events based on Generative Adversarial Network (GAN) is proposed. First of all, the ramping events get identified and separated from the database of historical wind power, and the feature set of historical ramping events is then extracted according to the waveform and meteorological factors. Taking the feature set which integrates similar feature with historical one as the input of GAN, the simulated ramping data are continuously produced through the adversarial training of the generator and discriminator, thus enriching the ramping database. After that, the expanded ramping database can be applied to predict the ramping power through the LSTM model. An experiment based on the wind power dataset in a certain area of northwest China further verifies the effectiveness and superiority of this method compared with traditional ones.