A novel method for controllable and interpretable load scenario generation based on LPR-Net and SRGAN

Accurate load scenario generation is critical for capturing complex demand uncertainty in power system planning under varying seasonal conditions. However, existing methods often lack controllability over scenario attributes and provide limited interpretability, making it difficult to understand the influence of load characteristic indicators on generated scenarios. These limitations hinder reliable decision-making in demand-side management. This study aims to develop a controllable and explainable framework for generating high-resolution load scenarios that capture complex peak-valley patterns. A joint probability model of daily load characteristic indicators is first established using Kernel Density Estimation and Gaussian copulas to capture their statistical dependencies. A two-stage generation framework is then introduced: (i) the Load Profile Reconstruction Network (LPR-Net) deterministically reconstructs low-resolution load profiles from sampled indicators, and (ii) a Super-Resolution Generative Adversarial Network (SRGAN) stochastically refines these profiles to high-resolution scenarios by introducing realistic variability. SHapley Additive exPlanations (SHAP) are further applied to quantify the contribution of each indicator, enhancing interpretability. The proposed approach achieves superior alignment with historical load distributions across all seasons, reconstructs peak-valley patterns with high fidelity, and exhibits stable convergence compared to single-stage generative models. It captures both overall load levels and intraday variability while preserving statistical dependencies among indicators, demonstrating robustness and generalizability under diverse operating conditions. The framework supports demand response and reserve scheduling by providing reliable, interpretable scenarios, enhances power system planning, and facilitates privacy-preserving synthetic data sharing. These capabilities improve the adaptability and reliability of decision-making in modern power systems.