Abstract
Traffic forecasting is the cornerstone of effective transportation planning, serving as the primary basis for determining road network structures, capacity requirements, and provincial corridor layouts. The accuracy of these models directly influences the scientific validity of infrastructure investment. Traditional "four-stage" models often suffer from a linear, one-way process that fails to account for the dynamic impact of congestion on travel behavior. This paper proposes an improved model that integrates travel time under congested conditions into the traffic distribution phase. By introducing a "circular forecast flow method," the authors establish a direct feedback loop between traffic distribution and assignment, creating a forecasting tool that more closely mirrors real-world highway dynamics.
Introduction
In civil engineering, traffic forecasting dictates where we build and how we scale our roads. Planners traditionally rely on the four-stage model: Generation (trip volume), Distribution (trip destination), Modal Split (vehicle type), and Assignment (route selection).
However, the traditional approach treats these steps as a sequence, ignoring how road performance affects driver choices. If a primary route is congested, the "impedance"—the cost or difficulty of travel—increases, which should logically change how traffic is distributed across the network. This paper addresses this limitation by refining the distribution phase to account for real-time congestion, ensuring that the model reflects the adaptive nature of human travel behavior.
Methodology
The researchers enhanced each stage of the model with a focus on iterative feedback and calibration:
- Traffic Generation: The model uses the elastic coefficient method, linking traffic growth to regional GDP. A six-step calibration process was implemented to ensure growth projections matched historical trends from 2004–2010.
- Traffic Distribution: Instead of simple distance-based calculations, the study employs a dual-constrained gravity model. It uses a composite Gamma function ($f(d_{ij}) = a \cdot d_{ij}^b \cdot e^{cd_{ij}}$) to represent impedance, allowing the model to mathematically capture the "friction" of long-distance travel more accurately than standard power-law models.
- Traffic Assignment: To simulate realistic path selection, the authors utilized a Stochastic User Equilibrium (SUE) model, solved via the Method of Successive Averages (MSA). This iterative algorithm refines flow estimates until the system reaches a realistic balance.
Data and Calibration
The study is grounded in a regional dataset from Zhejiang Province, China, covering 11 major internal regions (e.g., Hangzhou, Ningbo) and 5 neighboring provinces. Historical data from 2004 and 2010 were used to calibrate the model's baseline.
As illustrated in Figure 3, the composition of traffic on Zhejiang's highways shifted significantly during the calibration period. The proportion of cars in the traffic mix rose from 44.3% to 65.4%, while the relative share of trucks and coaches decreased. This shift is a critical input for the modal split phase of the model.
Figure 5 further highlights the underlying drivers of this change, showing a dramatic increase in private car ownership from 61.6% to 82.4% of the total vehicle fleet. These trends allowed the researchers to set accurate elastic coefficients for future projections.
Key Results and Insights
- Massive Volume Growth: The model predicts a surge in demand. For example, Hangzhou’s attraction volume is forecasted to grow from approximately 645,000 Passenger Car Units (pcu) per day in 2013 to nearly 2.5 million pcu/d by 2035.
- The "Slowing" Growth Curve: While absolute volumes continue to rise, the rate of growth is predicted to decelerate. In major hubs, growth is expected to drop from ~7.4% (2010-2015) to ~2.1% (2031-2035), reflecting a maturing economic landscape.
- Congestion Feedback: By considering travel time under congested conditions, the "circular forecast flow method" successfully identifies highway sections that will face the most pressure, providing a more reliable roadmap for long-term infrastructure planning.
Limitations and Future Work
While the model provides high-fidelity macro-level planning, it relies heavily on historical modal splits and GDP-linked growth. Future work could integrate "micro" simulations to analyze specific interchange designs. Additionally, the model does not yet account for disruptive technologies like autonomous trucking or the expansion of high-speed rail, which may decouple traffic growth from traditional economic indicators.
Summary updated