Scenario Generation Method for Emerging Source-Load Profiles Based on AVMD-DCGAN-GP

In light of the increasing complexity and uncertainty of modern power systems, this paper proposes an advanced generative model, AVMD-DCGAN-GP, for the realistic scenario generation of emerging source-load patterns such as photovoltaic (PV) output, traditional loads, and electric vehicle (EV) charging behaviors. The proposed model leverages Adaptive Variational Mode Decomposition (AVMD) to extract multi-scale temporal features from historical data, which are then used as conditional inputs for a Deep Convolutional Generative Adversarial Network (DCGAN). To improve training stability and mitigate mode collapse, a gradient penalty (GP) mechanism is incorporated into the loss function. Experimental results demonstrate that the AVMD-DCGAN-GP model achieves high fidelity in both statistical and temporal domains across various load types, offering a powerful tool for scenario-based analysis in power system planning, demand forecasting, and smart grid applications.