Accurate prediction of regional short-term load plays a crucial role in ensuring the stable operation of power systems and optimizing energy resource allocation. However，regional short-term load is influenced by a combination of various factors，and the electricity consumption characteristics of different user groups differ significantly. Traditional prediction methods exhibit shortcomings in terms of interpretability and accuracy. To address this，this paper proposes an explainable prediction method for regional short-term load based on multimodal influencing factors and user classification. Firstly，multimodal features are extracted from multiple dimensions such as date attributes，meteorological conditions，and socio-economic indicators. The label encoding method is then employed to convert these multimodal features into numerical labels，which serve as input features for subsequent load forecasting. Secondly，considering the differences in electricity consumption behaviors and load responses among user groups including industrial，commercial，agricultural，and residential users，a Bayesian optimization (Optuna)-based extreme gradient boosting (XGBoost) model is constructed to conduct load power prediction for each group separately. The regional total load is obtained by aggregating the prediction results of these four user categories. Finally，the SHapley Additive exPlanations (SHAP) method is introduced to analyze the contribution of various influencing factors to the load prediction and the interactions among different factors，thereby enhancing the model's interpretability. Validation using actual data from a region in Northwest China demonstrates that the proposed combined model achieves better predictive performance and higher forecasting accuracy.