随着全面建成小康社会的扎实推进，人民群众收入稳步增加，闲暇时间持续增多，健康水平不断提升，旅游业已迎来黄金发展。旅游业的飞速成长也给景区和经营管理者带来了新的挑战，大量游客短时聚集不仅会破坏游客的旅游体验，甚至还成为影响旅游安全的不稳定因素。为了提升旅游精细化和智能化管理水平，实现景区客流量实时预测，本文提出基于多层注意力超图神经网络的客流预测模型，为管理部门和景区提供科学便捷的决策依据，预防旅游安全事故等问题，为优质旅游发展提供支持，推动旅游业持续健康成长。本文的主要研究工作如下：（1）本文系统地研究和总结了国内外旅游景区客流预测的现状。针对目的地时空维度研究、客流预测影响因素研究、客流预测方法研究和基于机器学习的客流预测研究开展了详细的分析与总结。本文面向景区游客流量预测存在的问题，系统总结了利用网络数据和深度学习技术的研究现状和预测优势，为提出实时游客流量预测模型提供了理论支持。（2）本文系统研究了以小时为单位的实时客流量预测，利用景区客流量有关的海量异构时空数据，我们使用带有注意力机制的序列到序列改进模型来构建基于多层注意力超图神经网络的客流预测系统。本文将GeoMAN模型作为基本预测模型，引入超图神经网络对景区之间高阶关联关系进行建模、增加与景区客流量相关的外部数据输入，使数据集更具代表性；使用高斯分布来表征时间序列时刻结果，使预测结果具有更好的统计意义；为样本序列的极值设置评估和惩罚函数，提高预测模型的精度。实验证明，使用基于多层注意力超图神经网络模型进行实时客流量预测符合精准度要求 。（3）针对本文研究成果对旅游业的管理变革进行了深入探索。一是设计了景区流量监测演示系统，实现了景区流量实时监测预测动态展示，并配套制定了基于预测的景区高峰期应对措施，将传统的被动应对等待转变成主动应变管理，增强景区高峰期应对能力，提升旅游精准治理水平。二是设计了诸多应用场景，一方面为开放式景区加强客流量监测和预测打开了思路，依靠数据分析，无需进行传感器、摄像头等基础设施建设，可解决长期困扰该类型景区的问题；另一方面，该方法可提高各级旅游部门数据验证和管理能力，有效防止数据造假。

With the solid progress of building a well-off society in an all-round way, the income of the people has steadily increased, leisure time has continued to increase, and the health level has continued to improve. Tourism has ushered in golden development. The rapid growth of the tourism industry has also brought new challenges to scenic spots and operators. The short-term gathering of a large number of tourists will not only destroy the tourist experience of tourists, but also become an unstable factor affecting tourism safety. In order to improve the level of refined and intelligent management of tourism and realize the real-time prediction of passenger flow in scenic spots, this paper proposes a passenger flow prediction model based on multi-layer attention hypergraph neural network to provide scientific and convenient decision-making basis for management departments and scenic spots to prevent tourism safety accidents Provide support for the development of high-quality tourism and promote the sustained and healthy growth of the tourism industry.The main research work of this paper is as follows:(1) This paper systematically studies and summarizes the current situation of passenger flow forecasting in tourist attractions at home and abroad. Detailed analysis and summary are carried out for the research on the time and space dimensions of destinations, the research on the influence factors of passenger flow prediction, the research on passenger flow prediction methods and the research on passenger flow prediction based on machine learning. This article focuses on the problems of tourist flow prediction in scenic spots, systematically summarizes the research status and prediction advantages of using network data and deep learning technology, and provides theoretical support for proposing real-time tourist flow prediction models.(2) This paper systematically studies the real-time passenger flow forecast in hours. Using the massive heterogeneous spatiotemporal data related to the passenger flow of scenic spots, we use a sequence-to-sequence improvement model with an attention mechanism to build a multi-layer attention super The passenger flow prediction system based on graph neural network. In this paper, the GeoMAN model is used as the basic prediction model, the hypergraph neural network is introduced to model the high-order relationship between scenic spots, and the external data input related to the passenger flow of the scenic spots is increased to make the data set more representative; Gaussian distribution is used to represent Time series results make the prediction results have better statistical significance; set evaluation and penalty functions for the extreme values ?​of the sample series to improve the accuracy of the prediction model. Experiments have proved that the use of a multi-layer attention hypergraph neural network model for real-time passenger flow prediction meets the accuracy requirements.(3) Based on the research results of this article, the management reform of the tourism industry is deeply explored. The first is to design a flow monitoring demonstration system for scenic spots to realize the dynamic display of real-time flow monitoring and forecasting in scenic spots, and to formulate corresponding measures to respond to peak periods of scenic spots based on predictions, to transform the traditional passive response and waiting into active response management to enhance the peak period response of scenic spots Ability to improve the level of precision tourism management. The second is to design many application scenarios. On the one hand, it opens up ideas for enhancing passenger flow monitoring and forecasting in open scenic spots. By relying on data analysis, there is no need for infrastructure construction such as sensors and cameras, which can solve the problems that have plagued this type of scenic spots for a long time. On the one hand, this method can improve the data verification and management capabilities of tourism departments at all levels, and effectively prevent data fraud.