A new forecasting system for high-speed railway passenger demand based on residual component disposing
Introduction
Over the past decade, the extraordinary tempo of high-speed railway (HSR) development in several countries has attracted the continuous attention of the world. Especially in China, as a rapid inter-city transportation mode, HSR has reduced the heavy burdens of passenger transport for busy trunk lines among metropolises significantly [1]. Undeniably, there is a broad consensus that the accurate prediction of high-speed railway passenger flow is a significant issue for both construction and operation of the HSR system, such as planning and development of railway network, train operation plan, train or vehicle dispatching, and power supply operation management [2]. Aiming to develop relevant planning and control the system in a proactive manner, a predictive procedure has to be incorporated in the transportation system [3].
Many relevant items would be taken into consideration when the majority of travelers select trip modes, such as motivation travel, economic income, and personal preference. The passenger demand of HSR is usually affected by various factors [4]. Especially, in China, as the country with the largest population and the fastest development of high-speed railway, its high-speed railway passenger flow also has its own characteristics. Usually, the daily passenger demand during the weekend is generally greater than that of weekdays of the week, and the number of passengers during holidays and festivals outnumbers that of workday [5]. Thus the periodic behavior of travel characteristic would lead to the seasonal evolution hidden in passenger demand time series. From the perspective of development and evolution trend, in competition with aviation, common speed railway, and road infrastructure, HSR is favored by increasing travelers for a more comfortable and safer journey experience, which is the internal cause of the trend characteristics of HSR passenger flow. Both trend and seasonality are the main features of passenger demand data and are relatively easy to be captured by predictors. In the majority of cases, however, the uncertainty of individual trip mode is the main reason of chaotic and random presented in passenger demand data, which influences the forecasting procedure and results greatly. Thus, the chaotic and random property cannot be overlooked for accurate forecasting.
In most of the research work, the core work to achieve accurate prediction is how to deal with the complex characteristics of time series, such as random, chaotic, fluctuation and so on. To minimize the difficulty of prediction, a variety of data processing mechanisms have been developed to reduce the complexity of data and improve the prediction performance. In recent years, a kind of empirical mode decomposition (EMD) based on decomposition ensemble technology can effectively reduce the complexity of data, which has attracted extensive attention [6]. Its main technical principle is to decompose complex data into several relatively single modal components according to different spectra. Subsequently, its derivative methods and other similar treatment methods are also studied and developed, such as ensemble empirical mode decomposition (EEMD) [7], complementary ensemble empirical mode decomposition(CEEMD) [8], variational mode decomposition (VMD) [9], wavelet transform (WTB) [10], seasonal and trend decomposition using loess (STL) [11], etc. To be sure, by decomposing the complex data into several components with their own frequencies, the complexity of the data is greatly reduced, thereby improving the prediction accuracy significantly. Regrettably, the randomness hidden in the residual and high-frequency terms cannot be well mined and processed, which leaves the trouble and thorny obstacle to accurate prediction.
In the existing research, many scholars have paid attention to the analysis of residual terms, including quadratic decomposition strategy [12], error factor correction technology [13] and so on. However, excessive decomposition of high frequency terms may change the original characteristics of randomness. As far as we know, few studies investigate the remainder after the exclusion of structural components (trend and seasonality variations) from the perspective of residual component disposing. Whereas, in this study, it is believed that some predictive information remains in the remainder and if discarding them thoroughly would certainly exert a negative influence on prediction accuracy. The technical advantage of this idea is that it can truly describe the original appearance of randomness. The HSR passenger flow data involving individual travel behavior shows obvious randomness. When we separate the structural components from the original data, including the trend term and the period term, the remaining residual term covers the main random factors. It is easy to learn the evolution law hidden in the structural components except the residual term. How to effectively utilize the random factors contained in the residual is the key problem to obtain accurate prediction. To a great extent, we believe that the weak but indispensable information is still contained in the remainder component, and some random incidents occurred in the past may also happen in the future. Based on such a theoretical consideration, moving block bootstrap (MBB) is employed to bootstrap the residual series and generate a large number of similar samples, so as to simulate the stochastic feature of future series and improve the reliability of the data.
As a way to fill the gap in this issue, this paper develops a new hybrid system with deterministic and probabilistic forecasting capacities for HSR passenger demand. The data is firstly decomposed into trend, seasonality components, and residual by the seasonal and trend decomposition using loess (STL). Then, the residual is bootstrapped by moving block bootstrap (MBB), and the obtained information fragments is disposed to construct one thousand similar samples by virtue of extrapolating the disaggregated subseries, thereby simulating the stochastic feature of future series. After the extrapolation procedure, the generated similar deseasonalised (S) and detrended (T) subseries are predicted by seasonal autoregressive integrated moving average (SARIMA), kernel extreme learning machine (KELM), respectively. And support vector regression (SVR) is used in similar residuals prediction, including detrended series (TR) and deseasonalised series (SR). Similar global forecasts are obtained according to the reaggregation of the extrapolations. Finally, bagging predictors is performed to aggregate the obtained similar forecasts to get the final forecast results. To provide more profound advice for forethoughtful planners of railway operation department, we give deterministic and probabilistic forecasts of HSR passenger demand from one-step to three-step.
The rest of this paper is composed as follows. Sections Section 2 gives the literature review revolving around passenger demand of HSR forecasting. Sections Section 3 introduces the architecture of our established system and elucidates the detailed information of models and strategies adopted in this paper. In the 4th Section, the experiment setup and forecasts analysis are presented in detail. Finally, Section 5 represents the summary of this paper and further work.
Section snippets
Related forecasting technique
Aiming to assist transportation systems in planning and operating, including aviation, railway, freeway, and subway [14], [15], [16], [17], a slew of prediction techniques possessing distinctly different merits are available. All of these forecasting models can be classified as parametric model, non-parametric model and hybrid methods.
Parametric methods call for a group of mathematic or statistic equations with undetermined coefficients to simulate the law of data evolution [18]. In practice,
Methodology
In this section, we first expound the basic overview of our established architecture, and then discuss each individual component of the forecasting system in detail.
Experiment analysis
In this section, we first present an integrated feature analysis to original time series allowing us to select suitable pre-processing method for them. Afterward, we describe all the considered benchmark frameworks and evaluation metrics used in the assessment of established system. Finally, two case studies are conducted to expound the established system using real-world datasets of HSR passenger demand. The overall verification process is conducted on Matlab R2018a and E-views 10 environment
Conclusion
In this study, a hybrid system with learning mechanism simulating stochastic component of future data is presented for deterministic and probabilistic prediction. Several measures are of great importance for ensuring its efficiency and credibility. Initially, following STL decomposition, the MBB technique is employed to generate the similar irregular component. Furthermore, the residual possessing weak autocorrelation is involved in the predicting procedure by the principle of extrapolating the
CRediT authorship contribution statement
Wenxiu Cao: Conceptualization, Methodology, Software, Writing - original draft, Writing - review & editing. Shaolong Sun: Resources, Data curation, Investigation, Supervision. Hongtao Li: Resources, Formal analysis, Visualization, Funding acquisition, Validation.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This research is supported by the National Natural Science Foundation of China (Project No. 11361031), National Natural Science Foundation of China (Project No. 71761025), the grant from Scientific Research and Development Project of China Railway (N2020X015), the Science and Technology of Gansu Province Fund Project, China (Project No. 20JR5RA394), the Fundamental Research Funds for the Central Universities, China (Project No. xpt012020022), Beijing Natural Science Foundation, China (Project
References (56)
- et al.
Adaptive Kalman filter approach for stochastic short-term traffic flow rate prediction and uncertainty quantification
Transp. Res. Part C Emerg. Technol.
(2014) Forecasting demand for high speed rail
Transp. Res. Part A Policy Pract.
(2014)- et al.
A cross-boundary travel tale: Unraveling Hong Kong residents’ mobility pattern in shenzhen by using metro smart card data
Appl. Geogr.
(2021) - et al.
Forecasting the short-term metro passenger flow with empirical mode decomposition and neural networks
Transp. Res. Part C Emerg. Technol.
(2012) - et al.
A new secondary decomposition ensemble learning approach for carbon price forecasting
Knowl.-Based Syst.
(2021) - et al.
A novel hybrid decomposition-ensemble model based on VMD and HGWO for container throughput forecasting
Appl. Math. Model.
(2018) - et al.
A novel wavelet-SVM short-time passenger flow prediction in Beijing subway system
Neurocomputing
(2015) - et al.
A novel hybrid model based on STL decomposition and one-dimensional convolutional neural networks with positional encoding for significant wave height forecast
Renew. Energy
(2021) - et al.
A new secondary decomposition-ensemble approach with cuckoo search optimization for air cargo forecasting
Appl. Soft Comput. J.
(2020) - et al.
Multi-step ahead wind speed prediction based on optimal feature extraction, long short term memory neural network and error correction strategy
Appl. Energy
(2018)
Optimized configuration of exponential smoothing and extreme learning machine for traffic flow forecasting
IEEE Trans. Ind. Inform.
Short-term traffic flow rate forecasting based on identifying similar traffic patterns
Transp. Res. Part C Emerg. Technol.
Short-term forecasting of high-speed rail demand: A hybrid approach combining ensemble empirical mode decomposition and gray support vector machine with real-world applications in China
Transp. Res. Part C
Non-parametric hybrid models for wind speed forecasting
Energy Convers. Manag.
A novel hybrid decomposition-and-ensemble model based on CEEMD and GWO for short-term PM2.5 concentration forecasting
Atmos. Environ.
Forecasting with artificial neural networks: The state of the art
Int. J. Forecast.
Daily long-term traffic flow forecasting based on a deep neural network
Expert Syst. Appl.
Research and application of an innovative combined model based on a modified optimization algorithm for wind speed forecasting
Measurement
Comparative study on the time series forecasting of web traffic based on statistical model and generative adversarial model
Knowl.-Based Syst.
Short-term demand forecasting for online car-hailing using ConvLSTM networks
Phys. A Stat. Mech. Appl.
A network traffic forecasting method based on SA optimized ARIMA-BP neural network
Comput. Netw.
Multi-objective algorithm for the design of prediction intervals for wind power forecasting model
Appl. Math. Model.
A multi-objective wind speed and wind power prediction interval forecasting using variational modes decomposition based multi-kernel robust ridge regression
Renew. Energy.
Constructing prediction interval for artificial neural network rainfall runoff models based on ensemble simulations
J. Hydrol.
Mixed kernel based extreme learning machine for electric load forecasting
Neurocomputing
Urban traffic flow forecasting through statistical and neural network bagging ensemble hybrid modeling
Neurocomputing
A novel hybrid model for short-term high-speed railway passenger demand forecasting
IEEE Access
Prediction of high-speed railway passenger traffic volume based on integrated method
ICTE
Cited by (1)
Fast autoregressive tensor decomposition for online real-time traffic flow prediction
2023, Knowledge-Based Systems