Electric Vehicle Identification Model Based on Net Load Decomposition and Two-Stage Decision

To address the challenges of identifying electric vehicles (EVs) in user-side scenarios where multi-source load data is coupled with high-penetration distributed photovoltaics (PV), we propose a robust EV identification framework based on net load decomposition and a two-stage decision-making process. Initially, a context-aware source-supervised separation (CSSS) algorithm is employed to decouple PV output from the net load, effectively eliminating PV-induced interference by constructing targeted feature vectors. Subsequently, four key features characterizing EV charging behavior are extracted to feed into a hierarchical identification model. The first stage utilizes a Composite Charging Characteristic Index (CCCI) for rapid preliminary screening, while the second stage implements sample-adaptive weighted stacking ensemble learning for high-precision detection. Experimental results demonstrate that the proposed method achieves an identification accuracy of 96.33%, with the load decomposition stage contributing a 1.2% improvement. This framework provides a reliable technical foundation for load analysis and demand-side management in distribution networks with high PV integration.