Regional EV Charging Load Forecasting Based on SCLD and FCW

Against the backdrop of global energy transition and the continuous growth in electric vehicle (EV) market penetration, accurate forecasting of EV charging load is critically important for guaranteeing the safe and stable operation of power grids. Most existing forecasting approaches rely on artificial intelligence (AI) models trained with large-scale and continuous historical data, which imposes stringent requirements on the collection of EV charging load data. To address this issue, this paper proposes a novel method for EV charging load forecasting under small sample and discontinuous data conditions. Firstly, the differences between the daily load curves of EV charging are characterized by local dynamic time warping (LDTW) distance. And a spectral clustering algorithm based on LDTW distance (SCLD) is proposed to realize the classification of daily EV charging load patterns. Secondly, feature correlation weights (FCWs) derived from eXtreme gradient boosting (XGBoost) with one-hot encoding of input features are introduced to quantify the influences of features such as district-level attributes and weather conditions on daily EV charging load. Then, a method for determining the category of daily EV charging load based on FCWs and Hamming distance is put forward. On this basis, a daily EV charging load forecasting framework is established via weighted fitting of similar intra-class samples based on category judgment. Finally, to validate the effectiveness of the proposed method, a case study is carried out using EV charging load data and corresponding feature data of 62 typical days across 16 administrative districts in Shanghai from 2023 to 2025. The results demonstrate that the proposed method effectively addresses the challenging problem of EV charging load forecasting under small sample and discontinuous data conditions.