Knowledge and Data Dual-Driven Model for Plug-In Vehicle Load Identification

Large-scale integration of plug-in electric vehicles (PEVs) will impact the planning and operation of power systems. Accurately identifying PEVs serves as a crucial prerequisite for the establishment of charging station infrastructure. Existing methods for PEV identification can be divided into two categories, i.e., knowledge-driven (KD) models and data-driven (DD) models. The former suffers from the impact of explicit features extracted by subjective experience on identification performance, while the implicit features obtained by the latter are sensitive to the quality of recorded data. An interpretable knowledge and data dual-driven (IKD-DD) model is, therefore, proposed to integrate the robustness and learning abilities of two kinds of models by a mutual information-guided joint training process. A multiattention (MA) mechanism is, moreover, adopted to fuse the explicit and implicit features to enhance the discrimination ability of the feature set instead of conventional feature concatenation. Besides, a prototype-based contrastive loss function was introduced to learn PEV charging behavior even with limited labeled samples. Experimental results demonstrate that the proposed PEV load identification method achieves an identification accuracy exceeding 90% across various simulated and real datasets. Compared to existing methods, the proposed model shows significant improvement and robust performance, even when applied to datasets with limited labels and poor quality.