The Temporal Random Scene Generation Method Considering Source-Load Correlation

With the continuous increase in the penetration rate of renewable distribution generation (RDG) and the integration of various flexible controllable loads in new distribution systems, the uncertainty from both generation and demand sides poses significant challenges to the planning and scheduling of active distribution networks (AND). In response to this issue, this paper proposes a novel distribution network joint scenario generation method incorporating source-load correlation variational autoencoder (VAE), accompanied by the design of appropriate encoder and decoder network architectures. This method intrinsically integrates a position encoding module into the conventional VAE structure, employs temporal convolutional networks (TCN) for extracting correlation features of multidimensional stochastic variables related to source and load. Subsequently, simulation analysis is conducted based on actual photovoltaic and load data. The results demonstrate the effectiveness of the proposed model and method in learning the stochasticity and correlation of multi-dimensional time series variables, thus providing a reliable scenario modeling approach for planning and long-term operation.