基于改进Transformer模型的景区短时客流预测研究

doi:10.16381/j.cnki.issn1003-207x.2023.1927

Abstract

Abstract:

Current research on tourism demand forecasting primarily focuses on data with annual， monthly， and daily frequencies， leaving a gap in the study of short-term， high-frequency passenger flow predictions. A framework for predicting the flow of visitors to tourist attractions is studied based on an improved Transformer model. By collecting data every 15 minutes from February to August 2023 on seven 5A-level tourist sites in Beijing， including the Summer Palace， the Forbidden City， and the Temple of Heaven， etc， the Informer， Autoformer， and Fedformer models—three Transformer-based deep learning models are enhanced—using the TPE optimization algorithm to conduct one-step and multi-step predictions of high-frequency passenger flows in Beijing's seven scenic areas. These are then evaluated against the precision of other deep learning models （DeepAR， TCN， LSTM）， machine learning models （GBRT）， and time series models （ARIMA） across various forecasting scenarios. The performance of these models is evaluated using Mean Squared Error （MSE）， Mean Absolute Error （MAE）， and Improvement Rate （IR） metrics， along with DM tests for statistical significance. The results reveal that the three modified Transformer models consistently outperformed alternative methods， particularly the Informer model， which demonstrated superior accuracy across multiple prediction horizons， ranging from one step （15 minutes ahead） to 48 steps （one day ahead）. In the one-step scenario， the Informer achieved an average improvement rate of 14%， escalating to 23% and 40% for four and eight steps， respectively， and maintaining a substantial advantage even at the 48-step horizon， with an average improvement rate of 30%. These findings underscore the significant practical implications of high-frequency data and enhanced Transformer models in improving operational efficiency， managing congestion， optimizing visitor experiences， and providing decision-making information for tourist destinations， thereby promoting sustainable destination development. Furthermore， the body of research on tourism demand analysis and prediction models grounded is enriched in deep learning frameworks， thereby furnishing theoretical underpinnings and practical references instrumental to the advancement of theories and methodologies in tourism demand forecasting.

Key words: deep learning, Transformer, Informer, short-term tourist flow, tourism forecasting

CLC Number:

F590.8

Xin Li, Xu Zhang, Lean Yu, Shouyang Wang. Enhancing Short-term Tourist Flow Forecasting and Evaluation Using an Improved Transformer Framework[J]. Chinese Journal of Management Science, 2025, 33(2): 105-117.

Figures/Tables 14

方法	训练
方法	时间	内存
Informer	$O (L l o g L)$	$O (L l o g L)$
Autoformer	$O (L l o g L)$	$O (L l o g L)$
Fedformer	$O (L)$	$O (L)$
Transformer	$O (L 2)$	$O (L 2)$
GBRT	$O (L l o g L)$	$O (L l o g L)$
DeepAR	$O (L K 2)$	$O (L K 2)$
TCN	$O (L)$	$O (L)$
LSTM	$O (L l o g L)$	$O (L l o g L)$
ARIMA	$O (L 2)$	$O (L 2)$

景点	步长	Informer		Autoformer		Fedformer		GBRT		DeepAR		TCN		LSTM		ARIMA
景点	步长	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE
颐和园	1	0.098	0.221	0.1	0.222	0.102	0.222	0.172	0.315	0.112	0.238	0.254	0.399	0.111	0.234	0.754	0.824
	4	0.155	0.288	0.272	0.37	0.207	0.327	0.274	0.397	0.283	0.314	0.646	0.666	0.355	0.361	0.841	0.917
	8	0.277	0.354	0.426	0.497	0.377	0.454	0.435	0.501	0.309	0.424	1.264	0.931	0.536	0.449	0.751	0.866
	48	0.905	0.719	1.496	0.938	1.047	0.756	1.199	0.826	0.875	0.637	1.064	0.761	0.979	0.823	1.554	0.967
天坛	1	0.061	0.148	0.053	0.137	0.049	0.132	0.098	0.174	0.087	0.188	0.199	0.335	0.056	0.136	0.247	0.497
	4	0.149	0.245	0.207	0.301	0.15	0.246	0.261	0.302	0.218	0.309	0.504	0.529	0.164	0.277	0.231	0.481
	8	0.380	0.395	0.453	0.455	0.324	0.382	0.537	0.476	0.501	0.472	1.001	0.741	0.523	0.477	0.215	0.460
	48	1.227	0.806	1.478	0.856	1.263	0.776	2.37	1.071	1.342	0.835	1.349	0.791	1.281	0.788	0.343	0.578
故宫	1	0.031	0.115	0.032	0.121	0.026	0.102	0.038	0.12	0.035	0.116	0.087	0.216	0.039	0.119	0.752	0.867
	4	0.073	0.181	0.101	0.199	0.066	0.17	0.11	0.206	0.104	0.175	0.28	0.408	0.127	0.206	0.685	0.827
	8	0.122	0.230	0.414	0.442	0.138	0.26	0.226	0.305	0.145	0.203	0.595	0.603	0.228	0.249	0.593	0.768
	48	0.316	0.389	0.833	0.726	0.317	0.402	0.688	0.596	0.335	0.306	0.692	0.624	0.640	0.754	0.321	0.468
奥林匹克公园	1	0.097	0.157	0.102	0.164	0.101	0.16	0.118	0.18	0.101	0.158	0.129	0.211	0.120	0.161	0.131	0.277
	4	0.178	0.259	0.25	0.297	0.213	0.248	0.228	0.27	0.183	0.257	0.344	0.369	0.213	0.284	0.249	0.275
	8	0.211	0.313	0.398	0.402	0.329	0.341	0.341	0.364	0.305	0.368	0.55	0.528	0.288	0.287	0.372	0.386
	48	0.427	0.463	0.686	0.586	0.555	0.506	0.67	0.597	0.585	0.523	0.627	0.535	0.599	0.516	0.597	0.664
十三陵	1	0.018	0.095	0.019	0.095	0.02	0.098	0.027	0.111	0.022	0.098	0.033	0.139	0.028	0.112	0.022	0.111
	4	0.034	0.129	0.051	0.164	0.041	0.145	0.058	0.168	0.05	0.155	0.095	0.249	0.038	0.191	0.056	0.172
	8	0.056	0.169	0.087	0.227	0.073	0.199	0.106	0.234	0.084	0.205	0.184	0.352	0.061	0.217	0.103	0.221
	48	0.121	0.251	0.265	0.392	0.153	0.287	0.213	0.344	0.169	0.358	0.197	0.403	0.158	0.348	0.281	0.384
恭王府	1	0.081	0.183	0.084	0.185	0.084	0.17	0.097	0.189	0.082	0.184	0.183	0.308	0.165	0.245	0.101	0.123
	4	0.160	0.258	0.223	0.319	0.19	0.289	0.221	0.298	0.144	0.257	0.518	0.544	0.211	0.295	0.331	0.266
	8	0.219	0.314	0.416	0.454	0.317	0.389	0.402	0.413	0.323	0.391	1.006	0.785	0.325	0.345	0.368	0.315
	48	0.464	0.509	0.918	0.729	0.515	0.534	0.884	0.677	0.557	0.537	0.831	0.682	0.509	0.522	0.821	0.717
圆明园	1	0.007	0.053	0.009	0.065	0.008	0.058	0.010	0.068	0.009	0.064	0.011	0.077	0.017	0.095	0.011	0.108
	4	0.012	0.078	0.018	0.097	0.016	0.089	0.018	0.095	0.017	0.095	0.027	0.13	0.016	0.083	0.018	0.116
	8	0.017	0.098	0.026	0.122	0.021	0.107	0.026	0.121	0.026	0.123	0.043	0.166	0.022	0.126	0.027	0.147
	48	0.042	0.158	0.055	0.18	0.042	0.156	0.052	0.175	0.05	0.172	0.049	0.164	0.043	0.164	0.053	0.177

References 42

1	余乐安.基于人工智能的预测与决策优化理论和方法研究［J］. 管理科学， 2022， 35（1）： 60-66.
	Yu L A. Research on the theory and method of prediction and decision optimization based on artificial intelligence［J］. Journal of Management Science，2022，35（1）：60-66.
2	倪冬梅，赵秋红，李海滨.需求预测综合模型及其与库存决策的集成研究［J］.管理科学学报， 2013， 16（9）： 44-52+74.
	Ni D M， Zhao Q H， Li H B. Research on the integrated model of demand forecasting and its integration with inventory decision-making［J］. Journal of Management Sciences in China，2013，16（9）：44-52+74.
3	陈荣，梁昌勇，陆文星，等. 面向旅游突发事件的客流量混合预测方法研究［J］. 中国管理科学， 2017， 25（5）： 167-174.
	Chen R， Liang C Y， Lu W X， et al. The research of tourist flow hybrid forecasting model for tourism emergency events［J］. Chinese Journal of Management Science， 2017， 25（5）： 167-174.
4	尹隽，彭艳红，陆怡，等.基于深度神经网络的企业信息系统用户异常行为预测［J］. 管理科学， 2020， 33（1）： 30-45.
	Yin J， Peng Y H， Lu Y， et al. Prediction of abnormal behavior of enterprise information system users based on deep neural network［J］. Journal of Management Science， 2020，33（1）：30-45.
5	林宇，余元圆，张希，等.基于误差修正与深度强化学习的原油期货价格预测研究［J］.系统工程理论与实践，2023，43（1）：206-221.
	Lin Y， Yu Y Y， Zhang X， et al. Research on crude oil futures price prediction based on error correction and deep reinforcement learning［J］. Systems Engineering - Theory & Practice，2023，43（1）：206-221.
6	李洁，彭其渊，文超.基于LSTM深度神经网络的高速铁路短期客流预测研究［J］.系统工程理论与实践， 2021， 41（10）： 2669-2682.
	Li J， Peng Q Y， Wen C. Research on short-term passenger flow prediction of high-speed railway based on LSTM deep neural network［J］. Systems Engineering - Theory & Practice，2021，41（10）：2669-2682.
7	姚智胜，邵春福，熊志华. 基于小波包和最小二乘支持向量机的短时交通流组合预测方法研究［J］. 中国管理科学， 2007， 15（1）：64-68.
	Yao Z S， Shao C F， Xiong Z H. Research on short-term traffic flow combined forecasting based on wavelet package and least square support vector machines［J］. Chinese Journal of Management Science， 2007， 15（1）： 64-68.
8	潘和平，张承钊. FEPA-金融时间序列自适应组合预测模型［J］. 中国管理科学， 2018， 26（6）： 26-38.
	Pan H P， Zhang C Z. FEPA： An adaptive integrated prediction model of financial time series［J］. Chinese Journal of Management Science， 2018， 26（6）：26-38.
9	欧阳红兵，黄亢，闫洪举. 基于LSTM神经网络的金融时间序列预测［J］. 中国管理科学， 2020， 28（4）： 27-35.
	OuYang H B， Huang K， Yan H J. Prediction of financial time series based on LSTM neural network［J］. Chinese Journal of Management Science， 2020， 28（4）： 27-35.
10	陈海强，陈丽琼，李迎星，等.高频数据是否能改善股票价格预测？——基于函数型数据的实证研究［J］.计量经济学报， 2021， 1（2）： 426-436.
	Chen H Q， Chen L Q， Li Y X， et al. Can high-frequency data improve stock price prediction？An empirical study based on functional data［J］.China Journal of Econometrics Journal，2021，1（2）：426-436.
11	费兆奇，刘康. 中国宏观经济波动的高频监测研究——基于混频模型对日度经济先行指数的构建和分析［J］. 管理世界， 2019， 35（6）： 27-38.
	Fei Z Q， Liu K. Research on high-frequency monitoring of China’s macroeconomic fluctuations： Construction and analysis of daily economic leading index based on mixed frequency model［J］. Journal of Management World， 2019， 35（6）： 27-38.
12	Zhen H， Niu D， Wang K， et al. Photovoltaic power forecasting based on GA improved Bi-LSTM in microgrid without meteorological information［J］. Energy， 2021， 231： 120908.
13	Zheng W， Huang L， Lin Z. Multi-attraction， hourly tourism demand forecasting［J］. Annals of Tourism Research， 2021， 90： 103271.
14	武静，赵二龙，孙少龙，等. 多源异构数据驱动的后疫情时期旅游需求预测方法研究［J］. 计量经济学报， 2023， 3（2）： 350-366.
	Wu J， Zhao E L， Sun S L， et al. Research on tourism demand forecasting method in the post-epidemic period driven by multi-source heterogeneous data［J］. China Journal of Econometrics， 2023， 3（2）： 350-366.
15	Law R， Li G， Fong D K C， et al. Tourism demand forecasting： A deep learning approach ［J］. Annals of Tourism Research， 2019， 75： 410-423.
16	Silva E S， Hassani H， Heravi S， et al. Forecasting tourism demand with denoised neural networks［J］. Annals of Tourism Research， 2019， 74： 134-154.
17	李文，邓升，段妍，等. 时间序列预测与深度学习：文献综述与应用实例［J］. 计算机应用与软件， 2020， 37（10）： 64-70.
	Li W， Deng S， Duan Y， et al. Time series forecasting and deep learning： Literature review and application examples ［J］. Computer Applications and Software， 2020， 37（10）： 64-70.
18	徐小峰，余乐安，林姿汝，等. 基于特征融合的生鲜商品短期销量组合预测［J］. 管理科学学报， 2022， 25（12）： 102-123.
	Xu X F， Yu L A， Lin Z R， et al. Short-term sales portfolio prediction of fresh commodities based on feature fusion［J］. Journal of Management Sciences in China，2022，25（12）：102-123.
19	方雪清，吴春胤，俞守华，等. 基于EEMD-LSTM的农产品价格短期预测模型研究［J］. 中国管理科学， 2021， 29（11）： 68-77.
	Fang X Q， Wu C Y， Yu S H， et al. Research on short-term forecast model of agricultural product price based on EEMD-LSTM［J］. Chinese Journal of Management Science， 2021， 29（11）： 68-77.
20	Hewage P， Behera A， Trovati M， et al. Temporal convolutional neural （TCN） network for an effective weather forecasting using time-series data from the local weather station［J］. Soft Computing， 2020， 24： 16453-16482.
21	Sun S， Wei Y， Tsui K L， et al. Forecasting tourist arrivals with machine learning and internet search index［J］. Tourism Management， 2019， 70： 1-10.
22	Salinas D， Flunkert V， Gasthaus J， et al. DeepAR： Probabilistic forecasting with autoregressive recurrent networks［J］. International Journal of Forecasting， 2020， 36（3）： 1181-1191.
23	Han L， Ye H J， Zhan D C. The capacity and robustness trade-off： Revisiting the channel independent strategy for multivariate time series forecasting［J］. IEEE Transactions on Knowledge and Data Engineering， 2024， 36（11）： 7129-7142.
24	Cirstea R G， Micu D V， Muresan G M， et al. Correlated time series forecasting using multi-task deep neural networks［C］//Proceedings of the 27th ACM international conference on information and knowledge management， Torino， Italy， Ocober 22-26， Association for Computing Machinery， 2018： 1527-1530.
25	Rangapuram S S， Seeger M， Gasthaus J， et al. Deep state space models for time series forecasting［C］//Proceedings of the 32nd International Conference on Neural Information Processing Systems， Montréal， Canada， December 3-8 ， Curran Associates Inc，2018： 7796-7805.
26	Yan R， Liao J， Yang J， et al. Multi-hour and multi-site air quality index forecasting in Beijing using CNN， LSTM， CNN-LSTM， and spatiotemporal clustering［J］. Expert Systems with Applications， 2021， 169： 114513.
27	Li C， Zheng W， Ge P. Tourism demand forecasting with spatiotemporal features［J］. Annals of Tourism Research， 2022， 94： 103384.
28	Mao W， Zhu H， Wu H， et al. Forecasting and trading credit default swap indices using a deep learning model integrating Merton and LSTMs［J］. Expert Systems with Applications， 2023， 213： 119012.
29	高明，刘超，唐加福，等. 基于注意力神经网络的燃料电池寿命衰减预测［J］. 中国管理科学， 2023， 31（3）： 155-166.
	Gao M， Liu C， Tang J F， et al. Lifetime decay prediction of fuel cell based on attention neural network［J］. Chinese Journal of Management Science， 2023， 31（3）： 155-166.
30	吴洁，桂亮，刘鹏，等. 多维特征视角下基于图卷积网络的专利技术领域自动识别研究［J］. 中国管理科学， 2022， 30（12）： 185-197.
	Wu J， Gui L， Liu P， et al. Patent classification based on multi-dimensional feature and graph convolutional networks［J］. Chinese Journal of Management Science， 2022， 30（12）： 185-197.
31	Franceschi J Y， Dieuleveut A， Jaggi M. Unsupervised scalable representation learning for multivariate time series［C］//Proceedings of the 33rd International Conference on Neural Information Processing Systems， Vancouver BC， Canada， December 8-14 ， Curran Associates Inc， 2019： 4650-4661.
32	Devlin J， Chang M W， Lee K，et al. BERT： Pre-training of deep bidirectional transformers for language understanding［C］// Proceedings of 2019 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies， Minneapolis， Minnesota， June 2-7 ，Association for Computational Linguistics， 2019： 4171-4186.
33	Dong L， Xu S， Xu B. Speech-Transformer： A no-recurrence sequence-to-sequence model for speech recognition［C］// Proceedings of 2018 IEEE International Conference on Acoustics， Speech and Signal Processing （ICASSP）， Calgary， April 15-20， IEEE， 2018： 5884-5888.
34	Dosovitskiy A， Beyer L， Kolesnikov A， et al. An image is worth 16x16 words： Transformers for image recognition at scale［C］// Proceedings of the Ninth International Conference on Learning Representations， Virtual Conference， May 3-7 ， OpenReview， 2021.
35	Zhou H， Zhang S， Peng J， et al. Informer： Beyond efficient transformer for long sequence time-series forecasting［C］//Proceedings of the Thirty-Fifth AAAI Conference on Artificial Intelligence， Virtual Conference， February 2–9 ， Association for the Advancement of Artificial Intelligence， 2021： 11106-11115.
36	Wu H， Xu J， Wang J， et al. Autoformer： Decomposition transformers with auto-correlation for long-term series forecasting［J］. Advances in Neural Information Processing Systems， 2021， 34： 22419-22430.
37	Zhou T， Ma Z， Wen Q， et al. FEDformer： Frequency enhanced decomposed transformer for long-term series forecasting［C］//Proceedings of the 39th International Conference on Machine Learning， Baltimore， Marylan， July 17-23 ， Proceedings of Machine Learning Research， 2022： 27268-27286.
38	Bergstra J， Bardenet R， Bengio Y， et al. Algorithms for hyper-parameter optimization［C］//Proceedings of the 24th Advances in Neural Information Processing Systems， Granada， Spain， December 12-17 ， Curran Associates， Inc.， 2011： 2546-2554.
39	杨杰，冯芸，黄倩.基于MHPSO-NHMM-FIEGARCH-GED模型的Brent原油价格波动研究［J］. 中国管理科学， 2023， 31（6）： 265-275.
	Yang J， Feng Y， Huang Q. Research on brent crude oil price fluctuation based on MHPSO-NHMM-FIEGARCH-GED model［J］. Chinese Journal of Management Science， 2023， 31（6）： 265-275.
40	张歆悦，靳鹏，胡笑旋，等. 时间依赖型多配送中心带时间窗的开放式车辆路径问题研究［J］. 中国管理科学， 2024， 32（1）： 146-157.
	Zhang X Y， Jin P， Hu X X， et al. Research on the time-dependent multi-depot open vehicle routing problem with time windows［J］. Chinese Journal of Management Science， 2024， 32（1）： 146-157.
41	张婉莹，何耀耀，杨善林. 基于 TVFEMD-SE 和 YJQRG 的短期风电功率多步概率密度预测［J］. 系统工程理论与实践， 2022，42（8）： 2225-2242.
	Zhang W Y， He Y Y， Yang S L. Multi-step probability density prediction of short-term wind power based on TVFEMD-SE and YJQRG［J］. Systems Engineering - Theory & Practice， 2022，42（8）： 2225-2242.
42	Liu X， Liu A， Chen J L， et al. Impact of decomposition on time series bagging forecasting performance［J］. Tourism Management， 2023， 97： 104725.

模型	参数	选择范围	最优参数
Informer Autoformer Fedformer	seq_len	{1,2,6,12,24,48,96}	12
	label_len	{1,2,6,12,24,48,96}	6
	d_model	{256,512,1024}	512
	n_heads	{1,2,4,8,16}	8
	e_layers	{0,1,2,3,4,5}	2
	d_layers	{0,1,2,3,4,5}	1
	dropout	{0.1,0.5,0.01,0.05}	0.05
	batch_size	{8,16,32,64}	32
	learning_rate	{0.01,0.001,0.0001}	0.001

模型	1-step	4-step	8-step	48-step
Autoformer	-0.991	-1.787^**	-2.735^***	-2.186^**
Fedformer	-0.841	-1.118^*	-2.241^**	-1.558^**
GBRT	-2.405^**	-2.391^**	-2.736^**	-2.123^*
DeepAR	-1.232^*	-1.001^*	-1.714^**	-2.242^**
TCN	-1.998^**	-2.851^***	-3.667^***	-2.338^**
LSTM	-1.373^*	-1.483^**	-1.668^**	-1.291^*
ARIMA	-2.024^**	-1.889^**	-2.871^***	-1.187^*

[1]	Jiang-ze DU,Xi-zhuo CHEN,Le-an YU. Can Executives with IT Background Promote FinTech Innovation? [J]. Chinese Journal of Management Science, 2023, 31(12): 69-78.
[2]	Dong-xiao NIU,Zhuo-ya SIQIN,Dong-yu WANG,Xiao-min XU,Huan-fen ZHANG. Method and Application of Multi-value Chain Collaborative Data Mining in Power Equipment Enterprises Based on Deep Learning [J]. Chinese Journal of Management Science, 2023, 31(11): 321-331.
[3]	OUYANG Hong-bing, HUANG Kang, YAN Hong-ju. Prediction of Financial Time Series Based on LSTM Neural Network [J]. Chinese Journal of Management Science, 2020, 28(4): 27-35.
[4]	LI Ling, DING Shuai, LI Xiao-jian, YANG Shan-lin. Research on Smart Decision-Making Method for Upper Gastrointestinal Diseases Based on Electronic Gastroscopic Video [J]. Chinese Journal of Management Science, 2019, 27(11): 211-216.
[5]	KONG Fan-hui, LI Jian. Resilient Operation and Promotion Strategy of OEM Supply Chain under Supply Disruption Risk [J]. Chinese Journal of Management Science, 2018, 26(2): 152-159.

Enhancing Short-term Tourist Flow Forecasting and Evaluation Using an Improved Transformer Framework

RichHTML

PDF (PC)

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 14

References 42

Related Articles 5

Metrics

Comments

Recommended 0

模型	1-step	4-step	8-step	48-step
Autoformer	0.05	0.29	0.47	0.38
Fedformer	0.04	0.16	0.36	0.23
GBRT	0.18	0.22	0.38	0.36
DeepAR	0.04	0.03	0.31	0.27
TCN	0.25	0.48	0.62	0.32
LSTM	0.19	0.16	0.27	0.29
ARIMA	0.26	0.29	0.43	0.28
Average	0.14	0.23	0.40	0.30