Photovoltaic Random Scenario Prediction Based on Time Generative Adversarial Networks

Due to the randomness and time correlation of photovoltaic (PV) power output, the PV data obtained by conventional short-term point or interval prediction methods cannot support the robust scheduling and reliable operation of the power system under the condition of high proportion of PV integration. A photovoltaic random scenario prediction model based on Time - generative adversarial network (TimeGAN) is proposed. Firstly, the photovoltaic samples are decomposed into static and time-series data, and the network structures of embedding, reconstruction, generation and discrimination are constructed. Secondly, taking the static and time-series data as samples, the embedding and reconstruction networks are jointly trained. Finally, taking the output of the embedding network as samples, the generation and discrimination networks are trained adversarially, and the output of the generation network is fed back to the embedding-reconstruction training to test the fitting effect of the time series. Aiming to minimize the reconstruction, supervised and unsupervised losses, the model conducts in-depth learning of the photovoltaic time characteristics and predicts random scenarios. The validity of the proposed model is verified by the example calculation.