基于数字足迹与深度学习的城市旅游流量预测

Urban Tourism Flow Forecasting Based on Digital Footprints and Deep Learning

摘要: 随着智慧旅游的发展，如何实现对城市旅游流量的精准预测成为热点课题。本研究以杭州市为案例，结合2021~2024年百度指数关键词搜索数据与过夜旅客量数据，构建了一套基于数字足迹与深度学习的旅游预测体系。通过Spearman相关系数与随机森林算法，对关键词进行筛选与排序，选出8个与游客量高度相关的搜索词。在此基础上，构建BP神经网络与多特征LSTM神经网络模型，分别对游客量进行预测，并与传统ARIMA模型进行对比分析。研究结果显示，LSTM模型在拟合精度与波动趋势捕捉方面表现最优，MAPE最低为0.099，优于其他模型，验证了融合数字搜索数据与深度学习算法的有效性。研究为智慧旅游背景下的游客行为理解与资源调度优化提供了方法参考与实践支撑。

Abstract: With the advancement of smart tourism, accurate forecasting of urban tourist flows has become a critical research topic. This study takes Hangzhou as a case city and constructs a tourism flow prediction framework based on digital footprints and deep learning by integrating Baidu Index keyword search data and overnight tourist volume data from 2021 to 2024. Using Spearman correlation analysis and random forest algorithms, eight keywords most closely related to tourist volume were selected. Based on these, a BP neural network and a multi-feature LSTM neural network model were developed to predict tourist flows, and their performance was compared with a traditional ARIMA model. Results show that the LSTM model achieved the best performance in terms of fitting accuracy and trend capturing, with the lowest MAPE of 0.099, outperforming the other models. This study demonstrates the effectiveness of integrating digital search data and deep learning algorithms, and provides methodological and practical support for understanding tourist behavior and optimizing resource allocation under the smart tourism paradigm.