【目的】针对电动汽车多充电站在城市路网中布点分散、相互耦合导致负荷预测难度大的问题，提出一种融合路网、气象和日期特征的联合预测模型，旨在提高多充电站负荷预测的精度和效率，为新型电力系统的经济调度和高效运行提供技术支持。【方法】采用改进的迪杰斯特拉算法构建充电站之间的空间关联网络，通过图卷积神经网络（graph convolutional network， GCN）融合路网特征，气象变量通过斯皮尔曼秩相关系数筛选关键特征，日期特征以二进制编码表示工作日与非工作日的差异，并引入基于Mamba的选择性状态空间模型（selective state space model， SSSM）以捕捉长期时序特征并降低Transformer模型在长期预测中的高计算复杂度。【结果】以吉林省某市45个充电站的实际运行数据为样本进行验证，结果表明GCN-Mamba模型能够有效捕捉充电站之间的相互影响以及气象、日期等外部因素的作用，其均方根误差和平均绝对误差较传统模型显著降低，计算时间相比Transformer模型减少约30%，展现了优越性能。【结论】GCN-Mamba模型在电动汽车多充电站负荷预测中表现出高精度和高效性，创新性地融合路网空间关联和选择性状态空间模型，为新型电力系统负荷预测提供了新思路；但研究样本数据局限于单一城市，未来可扩展至多城市、多场景验证，并优化模型对极端天气条件的适应性。

[Objective] To address the challenge of load forecasting for multiple spatially dispersed and mutually coupled electric vehicle charging stations in urban road networks， a joint forecasting model integrating road network， meteorological， and data features is proposed. This model aims to enhance the accuracy and efficiency of multistation load forecasting， providing technical support for the economic dispatch and efficient operation of new power systems （NPSs）. [Methods] The model employs an improved Dijkstra algorithm to construct a spatial correlation network among charging stations and utilizes a graph convolutional network （GCN） to integrate road network features. Meteorological variables were selected using the Spearman rank correlation coefficient， and date features were binary-encoded to distinguish between weekdays and non-weekdays. To mitigate the high computational complexity of transformer models in long-term forecasting， a selective state-space model （SSSM） based on Mamba was introduced to capture long-term temporal features while reducing computational demands. [Results] The model was validated using operational data from 45 charging stations in a city in the Jilin Province. The results demonstrate that the GCN-Mamba model effectively captures inter-station interactions and the impacts of external factors such as meteorology and dates. Its root mean square error （RMSE） and mean absolute error （MAE） were significantly lower than those of traditional models， and its computation time was reduced by approximately 30% compared with transformer models， indicating superior performance in multistation load forecasting. [Conclusions] The GCN-Mamba model innovatively combines road network spatial correlations with the SSSM and offers a novel approach for NPS load forecasting. Furthermore， it exhibits high accuracy and efficiency. However， this study is limited by the use of a single-city dataset. Future work should include multi-city and multi-scenario validations， as well as improvements to the adaptability of the model under extreme weather conditions.