【摘要】 针对当前交通领域日客运量预测模型对节假日多维特征利用不足的问题，提出一种融合多维特征增强模块（MFE）的MFE-Informer模型。将节假日类型、持续时间及前后时段等信息映射为多维特征，通过MFE模块与多头注意力层相结合，使模型能够自适应聚焦节假日特征变化，从而增强对节假日影响及长期趋势的建模能力；以2019年某机场日客运量数据为例进行日客运量预测。研究结果表明：相较于长短时记忆神经网络（LSTM）模型、Informer及带二元标签的Informer等基线模型，MFE-Informer模型在相关性系数（R²）、均方误差(E_MS)、平均绝对误差(E_MA)和均方根误差(E_RMS)等指标上均表现更优；选取3类典型出行场景（清明假期、旺季某5d、国庆假期）进行验证，MFE-Informer模型均保持了较小的预测误差与更快的趋势响应能力，证明了该模型在不同出行模式下的稳健性。更多还原

【Abstract】 Aiming at the problem of insufficient utilization of multi-dimensional features of holiday in current daily passenger volume prediction models in the transportation sector, an MFE-Informer model that incorporated multi-dimensional feature enhancement（MFE） module was proposed, in which the information such as holiday type, duration, and surrounding time periods was mapped into multi-dimensional features. By integrating the MFE module with the multi-head attention layer, the proposed model could adaptively focus on feature variations of holidays, thereby enhancing its capability to model the impact of holidays and long-term trends. Taking the daily passenger volume data from a certain airport in 2019 as an example. The daily passenger valume forecasting was carried out. The research results show that compared to baseline models such as long short-term memory（LSTM） model, Informer and Informer with binary labels, the MFE-Informer model performs better in terms of evaluation metrics such as correlation coefficient（R²）, mean squared error(E_MS), mean absolute error(E_MA), and root mean squared error(E_RMS). Furthermore, three typical travel scenarios（Tomb-Sweeping Day, a certain 5-day period during peak season, and National Day holiday） are selected for validation. The MFE-Informer model maintains smaller prediction errors and faster trend response capability, which confirms the robustness of the proposed model across different travel patterns.更多还原