LineGen: Physics-Guided Super-Resolution Load Data Generation via A Straight-Line Path

Super-resolution load data (SRLD) is crucial for the reliability of distribution power systems. However, current distribution power systems struggle to support the extensive collection and transmission of SRLD. Utilizing deep learning models to directly generate SRLD from low-resolution load data (LRLD) remains challenging due to limited fitting capabilities and inherent poor interpretability. To address these challenges, this paper proposes a novel approach that establishes a data distribution transformation model (DDTM) and introduces LineGen, a physics-guided SRLD generation method. Unlike classical end-to-end deep learning models, LineGen initially constructs a transformation trajectory from LRLD to SRLD using straight-line ordinary differential equations, thereby theoretically implementing data distribution transformation. Subsequently, a time-embedded U-Net-enabled estimator is used to estimate the transformation rate at any time section along the straight-line transformation trajectory. More specifically, under the guidance of the DDTM, the proposed U-Net is trained iteratively with predict loss and correct loss, leading to the introduction of the predict-correct (PC) loss relay training method. Finally, SRLD is efficiently generated from LRLD via the straight-line path using a one-step Euler generation method. The physics-guided mechanism enables the highly accurate generation of SRLD with an enhancement of 1000-fold data resolution, which was previously considered beyond reach, while providing traceable physical interpretability. Comparative results with state-of-the-art approaches validate the high accuracy and efficiency of the proposed method.