A Data-Driven and Deep Learning-Based Method for Power System Operating Mode Identification

Efficient and accurate analysis to identify future power system operating modes is crucial for handling power system operation, planning, and stability analysis. This paper proposes a data-driven and deep learning-based method for analyzing typical operating modes in power systems, while also addressing the needs for identifying future operating modes. Firstly, an operating mode analysis method for power systems is developed based on an adaptive threshold Affinity Propagation (AP) clustering method with Dynamic Time Warping (DTW) that incorporating historical load sequences. The method employs a modified distance function and adaptive thresholds for clustering. Secondly, a day-ahead load forecasting method based on a Parallel Temporal Fusion Network (PTFN) model with temporal feature projection is proposed to address load uncertainty. Finally, based on the results of historical operating mode analysis and future load forecasting, a SHapley Additive exPlanation combined with Parallel Temporal Convolution Network embedded with the Squeeze-Excitation mechanism (SHAP-PTCN-SE) is proposed for identifying future operating modes of power systems. Numerical examples demonstrate the efficiency and accuracy of the proposed method in identifying future operating modes, providing guidance for system monitoring and protection.