【摘要】 为优化食堂资源配置与服务管理，本文提出了一种结合双向门控循环单元(Bidirectional Gated Recurrent Unit, BiGRU)和注意力机制(Attention)的组合模型，用于预测食堂人流量。该模型通过BiGRU提取历史数据中的时间依赖关系，再借助Attention机制对关键信息进行筛选，挖掘深层次特征，进一步降低多步预测误差。通过对某高校一卡通消费数据实验表明，BiGRU-Att模型在预测高校食堂窗口人流量任务上表现出卓越的性能。相比单一的BiGRU和支持向量回归(Support Vector Regression, SVR),其预测结果的均方误差分别降低了0.051和1.134,R~2分别提高0.017和0.374。研究证明，BiGRU-Att模型充分融合了循环神经网络的序列建模能力和注意力机制的特征权重自适应学习能力，更精确地捕捉人流量的动态演变规律，有助于提高餐饮行业科学决策能力，优化其服务管理流程。更多还原

【Abstract】 In order to optimize cafeteria resource allocation and service management, this paper proposes a combined model combining Bidirectional Gated Recurrent Unit(BiGRU) and Attention mechanism to predict cafeteria pedestrian flow. The model uses BiGRU to extract the time dependence relationship into the historical data, and then uses the Attention mechanism to screen the key information, mine the deep features, and further reduce the multi-step prediction error. Experiments on the consumption data of a university card in Zhejiang Province show that the BiGRU-Att model shows excellent performance in predicting the pedestrian flow of university canteen Windows. Compared to the single BiGRU and Support Vector Regression(SVR), it reduces the mean square error by 0.051 and 1.134, respectively, and increases the R-squared value by 0.017 and 0.374. Studies show that the BiGRU-Att model fully integrates the sequence modeling ability of recurrent neural network and the feature weight adaptive learning ability of attention mechanism, which can more accurately capture the dynamic evolution law of pedestrian flow, and help to improve the scientific decision-making ability of the catering industry and optimize its service management process.更多还原