Source-Load Scenario Generation Method Based on Temporal Generative Adversarial Network

To address the challenges of high variable dimensionality and complex spatiotemporal characteristics in monthly source-load power modeling, this paper proposes a temporal generative adversarial network-based framework for monthly source-load scenario generation. Firstly, an RV coefficient-based clustering technique is employed to extract representative daily states of the source-load profiles, while a Markov chain is used to characterize the state transition patterns of the source-load. Then, a temporal generative adversarial network is constructed by incorporating multi-head attention mechanisms and temporal convolutional networks, aiming to capture the temporal dynamics and spatial correlations of intraday source-load power. Furthermore, a monthly source-load power scenario generation method is developed using stochastic techniques. Finally, the generated source-load scenarios are evaluated using a comprehensive evaluation index system.