Hybrid CSA optimization with seasonal RVR in traffic flow forecasting

Zhangguo Shen, Wanliang Wang, Qing Shen, Zechao LiKSII Trans. Internet Inf. Syst 2017 · DOI

Abstract

Accurate traffic flow forecasting is critical to the development and implementation of city intelligent transportation systems. Therefore, it is one of the most important components in the research of urban traffic scheduling. However, traffic flow forecasting involves a rather complex nonlinear data pattern, particularly during workday peak periods, and a lot of research has shown that traffic flow data reveals a seasonal trend. This paper proposes a new traffic flow forecasting model that combines seasonal relevance vector regression with the hybrid chaotic simulated annealing method (SRVRCSA). Additionally, a numerical example of traffic flow data from The Transportation Data Research Laboratory is used to elucidate the forecasting performance of the proposed SRVRCSA model. The forecasting results indicate that the proposed model yields more accurate forecasting results than the seasonal auto regressive integrated moving average (SARIMA), the double seasonal Holt-Winters exponential smoothing (DSHWES), and the relevance vector regression with hybrid Chaotic Simulated Annealing method (RVRCSA) models. The forecasting performance of RVRCSA with different kernel functions is also studied.

The Challenge of Urban Traffic Flow

Predicting urban traffic flow is notoriously difficult because the data is both highly nonlinear and deeply seasonal. While standard models like ARIMA are common, they often struggle to capture "fast multilateral movements"—the sudden spikes and drops during rush hours.

Figure 1: Typical traffic flow data exhibiting strong seasonal patterns and nonlinear fluctuations over time.

Machine learning approaches like Support Vector Machines (SVM) have shown promise, but they can be computationally heavy for real-time systems. Furthermore, these models are highly sensitive to "hyper-parameters" (settings that define how the model learns), and finding the optimal settings is often a game of trial and error. This paper seeks to bridge the gap by creating a model that is sparse enough for real-time use, self-optimizing, and specifically tuned for the "seasonal" rhythms of city life.

Methodology: The SRVRCSA Model

The proposed SRVRCSA model integrates three distinct pillars to handle the complexities of traffic data:

  1. Relevance Vector Regression (RVR): Unlike the more famous SVM, RVR is a sparse Bayesian model. It provides the same functional form as an SVM but typically uses far fewer basis functions, making it faster and more suitable for real-time intelligent transportation systems.
  2. Chaotic Simulated Annealing (CSA): To find the best parameters for the RVR's kernel function, the authors use Simulated Annealing (SA). By injecting chaos—specifically using a logistic map to generate search variables—the algorithm maintains better randomness and uniformity, allowing it to "jump out" of local traps to find the true global optimum.
  3. Seasonal Adjustment: Recognizing that traffic on a Monday morning looks nothing like traffic on a Sunday afternoon, the authors implement a two-step seasonal index calculation to adjust the RVR output for specific peak periods.

Datasets and Experimental Setup

The study utilized real-world traffic flow data from The Transportation Data Research Laboratory (TDRL) at the University of Minnesota Duluth. Data was collected from March 30 to April 26, 2015, across six distinct road sections in the Minnesota Twin Cities and Rochester area (e.g., TH252SB, I-394 CDWB).

Figure 2: Comparative traffic flow profiles from different road sections, illustrating varying scales and intensities of peak-hour demand.

The data was captured at hourly granularity during morning (6:00–11:00) and evening (15:00–20:00) peaks. To ensure accuracy, the data was strictly separated into "weekday" and "weekend" categories to train and forecast separately.

Key Results and Performance

The SRVRCSA model was compared against three major baselines: SARIMA, DSHWES (Double Seasonal Holt-Winters), and a standard RVRCSA without seasonal adjustment. Using Root Mean Square Error (RMSE) as the primary metric, the results were conclusive:

  • Weekday Performance: SRVRCSA achieved an RMSE of 2.49, significantly beating SARIMA (4.66) and DSHWES (3.61).
  • Weekend Performance: The model was even more accurate here, with an RMSE of 1.85, compared to SARIMA (3.41) and DSHWES (3.29).
  • Kernel Sensitivity: Testing six different kernel functions revealed that the Gaussian Function (GF) consistently provided the best balance of accuracy across different road types.

Conclusion

The SRVRCSA model demonstrates that combining sparse Bayesian learning with chaotic optimization and seasonal adjustments provides a robust framework for urban traffic forecasting. By nearly halving the error rate of traditional time-series models, this approach offers a viable path forward for real-time intelligent transportation systems. Future work will look to integrate multivariate factors such as weather and lane occupancy to further refine predictive accuracy.

Summary updated

BibTeX
@article{shen2017hybrid,
  author = {Zhangguo Shen and Wanliang Wang and Qing Shen and Zechao Li},
  title = {{Hybrid CSA optimization with seasonal RVR in traffic flow forecasting}},
  journal = {KSII Trans. Internet Inf. Syst.},
  year = {2017},
  doi = {10.3837/tiis.2017.10.011},
}
Figure