Interval prediction of short-term traffic speed with limited data input: Application of fuzzy-grey combined prediction model
Introduction
Nowadays, urban traffic congestion has attracted attention for the operation of transportation systems around the world. In targeting this issue, many solutions have been provided to alleviate traffic congestion, where proactive systems are very effective (Cheng, Gau, Huang, & Hwang, 2012). Specifically, for proactive systems, a highly accurate prediction of short-term traffic parameters (i.e., flow, speed, occupancy, travel time) is a critical component. Compared with other traffic parameters, traffic speed can more directly reflect the traffic operation status. Moreover, the accurate prediction of short-term traffic speed is more helpful for implementing real-time traffic proactive control strategy (Pan, Sumalee, Zhong, & Indra-payoong, 2013). Past literatures on short-term traffic speed prediction are level prediction, which can only directly predict the average value of the traffic speed based on a series of historical data. However, the performance and accuracy of traffic speed level prediction often be influenced by various uncertainty factors such as weather, incidents, and driver behaviors. Thus, we need a forecast result that can reflect the uncertainty in traffic speed. Specifically, the prediction interval (PI) result can just deal with the uncertainty of traffic speed. Moreover, the accurate prediction interval (PI) of traffic speed can provide more useful information for traffic managers to make reasonable decisions. Therefore, it is also considered a significant concern for the development and implementation of the proactive traffic control systems in the future.
In general, most of the literatures on traffic speed prediction focuses on level prediction, and there has been little research on PI over the past decade. However, more and more scholars have studied the PI of traffic parameters recently. The conventional approaches for PI can be classified into two categories. One is that the PI can be obtained by using the analytical expression to calculate the variance or standard error of variables. For example, the generalised autoregressive conditional heteroscedasticity (GARCH) model, as one of the primary methods, is to obtain the PI results according to the variance error of variables, where the variance error of the variables can be obtained based on the probability density of level prediction results (Andersen, Bollerslev, Christoffersen, & Diebold, 2006). The other is that the upper and lower bounds of PI can be directly obtained by using the lower/upper bound estimation (LUBE) method (Quan et al., 2014).
It is necessary to point out that the above PI methods have higher requirements on the data sample size. However, it is difficult to get large amounts of data for some countries and regions, such as India, due to automated sensors and associated database are being implemented or damaged (Badhrudeen, Raj, & Vanajakshi, 2016). Moreover, such PI methods for short-term traffic parameters have not been developed or tested rigorously with limited data input. Thus, a PI method of short-term traffic parameters based on less data demanding would be more attractive. The grey prediction theory proposed by Deng (1989) provides a way to generate PI under limited data condition. The principle of the grey prediction interval algorithms is to first divide the original data series into upper and lower interval data series, then the PI results can be obtained by forecasting the upper and lower data series respectively. For instance, the grey straight horn band (GPBI) model proposed by Liu and Lin (2006) uses the general straight line to produce the upper and lower data series, and the grey interval prediction model proposed by Chen and Liu (2017) constructed a linear regression function to divide the original data series into two groups. These grey PI models select a simple method of data classification, which will cause a loss of information in the whitenization process. In this way, the results of PI for short-term traffic parameters are less reliable.
Currently, fuzzy theory provides a way of conducting the range values of data series. In this regard, fuzzy information granulation (FIG) is widely used for generating upper and lower interval data series, and this method also can maintain the integrity of the original data without losing any data information (Lu, Chen, Pedrycz, Liu, & Yang, 2015). Therefore, the FIG method can be used to generate the dispersion range of short-term traffic speed data series. The objective of this study is to propose a short-term traffic speed forecast interval system based on the FIG method and the grey prediction model with limited input data. However, the traditional grey prediction model cannot predict the traffic speed series very well because of the random fluctuation features of the traffic speed series. Furthermore, the model does not consider the characteristics that the current traffic speed is generally affected by past traffic conditions. Thus, this paper applies a grey autoregressive model (GARM) to predict the traffic speed, which adds an autoregressive mechanism based on the grey model. In order to verify the effectiveness of the proposed FIG-GARM model for prediction interval of traffic speed time series, the real-world traffic speed data from an urban freeway corridor in Edmonton are taken as the experimental data. The interval-grey model first order single variable (GM (1,1)), FIG-GM (1,1) model, and interval-GARM model are used as comparison models to investigate the PI performance for traffic speed series by calculating the performance indicators, including prediction interval coverage probability (PICP), average width interval (WI), and index P.
The rest of this paper is organized as follows. Section 2 introduces the briefly literature; the proposed model system is described in Section 3; Section 4 gives an empirical case to investigate the PI performance of the proposed model; finally, Section 5 proposes conclusions.
Section snippets
Literature review
This section summarizes the state-of-the-art of short-term traffic condition prediction models, short-term traffic condition interval prediction models and grey interval prediction models.
Proposed interval prediction framework
In this section, the proposed FIG-GARM model for short-term traffic speed is described below. It primarily introduces the FIG method, the GARM model and the PI framework for short-term traffic speed based on the FIG-GARM model.
Data description
Traffic speed data used in this paper were collected by the vehicle detection stations (VDS) from a freeway corridor in the city known as Whitemud Drive in Edmonton, Canada. The west to east direction segment between 170th street to 122th street was selected, and the selected corridor of the urban freeway was divided into nine segments based on the location of the detector as shown in the Fig. 2 (http://www.openits.cn/openData1/700.jhtml). The selected period is the morning peak period on
Conclusion
Over the decades, short-term traffic speed prediction, including level and interval prediction, has received a sustained attention from many transportation engineers and researchers for developing proactive traffic control and management systems. Interval prediction of traffic speed is more useful in making informed decisions than level prediction, because level prediction becomes less persuasive as traffic speed data are distributed more discretely. So, to obtain the interval forecasting
CRediT authorship contribution statement
Zhanguo Song: Writing - original draft, Methodology, Software. Wei Feng: Visualization, Investigation. Weiwei Liu: Data curation.
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.
Acknowledgement
This research was funded by the China Scholarship Council (No. 201906090213) and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (No. KYCX18_0151).
References (44)
- et al.
Volatility and Correlation Forecasting
Handbook of Economic Forecasting
(2006) Generalized autoregressive conditional heteroskedasticity
Journal of Econometrics
(1986)- et al.
Short-term freeway traffic parameter prediction: Application of grey system theory models
Expert Systems with Applications
(2016) - et al.
Advanced formation and delivery of traffic information in intelligent transportation systems
Expert Systems with Applications
(2012) - et al.
A spatiotemporal correlative k-nearest neighbor model for short-term traffic multistep forecasting
Transportation Research Part C: Emerging Technologies
(2016) - et al.
Memory properties and fractional integration in transportation time-series
Transportation Research Part C: Emerging Technologies
(2009) - et al.
Using interval information granules to improve forecasting in fuzzy time series
International Journal of Approximate Reasoning
(2015) - et al.
Real-time road traffic prediction with spatio-temporal correlations
Transportation Research Part C: Emerging Technologies
(2011) - et al.
Long short-term memory neural network for traffic speed prediction using remote microwave sensor data
Transportation Research Part C: Emerging Technologies
(2015) - et al.
Construction of prediction intervals using adaptive neuro-fuzzy inference systems
Applied Soft Computing Journal
(2018)
Deep learning for short-term traffic flow prediction
Transportation Research Part C: Emerging Technologies
Bayesian committee of neural networks to predict travel times with confidence intervals
Transportation Research Part C: Emerging Technologies
Short-term traffic speed forecasting hybrid model based on chaos-wavelet analysis-support vector machine theory
Transportation Research Part C: Emerging Technologies
The design of fuzzy information granules: Tradeoffs between specificity and experimental evidence
Applied Soft Computing Journal
Prediction interval of wind power using parameter optimized Beta distribution based LSTM model
Applied Soft Computing Journal
A hybrid short-term traffic flow forecasting method based on spectral analysis and statistical volatility model
Transportation Research Part C: Emerging Technologies.
Short-term prediction of traffic parameters-performance comparison of a data-driven and less-data-required approaches
Journal of Advanced Transportation
The Use of Subseries Methods for Estimating the Variance of a General Statistic From a Stationary Time Series
Annals of Statistics
Predictions of freeway traffic speeds and volumes using vector autoregressive models
Journal of Intelligent Transportation System
Neural-network-based models for short-term traffic flow forecasting using a hybrid exponential smoothing and Levenberg-Marquardt algorithm
IEEE Transactions on Intelligent Transportation Systems
Time series interval forecast using GM(1,1) and NGBM(1, 1) models
Soft Computing.
Uncertainty in traffic forecasts: Literature review and new results for the Netherlands
Transportation
Cited by (19)
A novel partial grey prediction model based on traffic flow wave equation and its application
2024, Engineering Applications of Artificial IntelligenceImproving conformalized quantile regression through cluster-based feature relevance
2024, Expert Systems with ApplicationsA fundamental diagram based hybrid framework for traffic flow estimation and prediction by combining a Markovian model with deep learning
2024, Expert Systems with ApplicationsA novel grey prediction model based on tensor higher-order singular value decomposition and its application in short-term traffic flow
2023, Engineering Applications of Artificial IntelligenceDistributional prediction of short-term traffic using neural networks
2023, Engineering Applications of Artificial IntelligenceA grey breakpoint prediction model and its application in forecasting and policy evaluation
2023, Engineering Applications of Artificial Intelligence