Source-linked AI summary
Langevin method for a continuous stochastic car-following model and its stability conditions
D. Ngoduy, S. Lee, M. Treiber, M. Keyvan-Ekbatani, H. L. Vu
TL;DR
Most car-following models are deterministic and do not represent uncertainty in human perception and driving behavior, which is associated with traffic oscillations. The paper uses coupled Langevin equations with an extended CIR process and derives stochastic linear stability conditions. Its analysis indicates stochasticity can destabilize low-speed traffic even when the deterministic model is stable.
Problem
Most car-following models are deterministic and do not capture uncertainty in human perception and driving behavior, factors linked to the formation and propagation of stop-and-go waves.
Source: §1 · pp. 1–2Method
The authors use coupled Langevin equations with an extended CIR process to model stochastic driver perception and derive stochastic linear stability conditions.
Source: §1 · p. 3Results
The analysis finds that stochasticity can destabilize traffic at low speeds even when the deterministic counterpart is stable.
Source: §1 · p. 3Takeaways & Limitations
The model offers a way to analyze how stochastic human behavior relates to traffic instability while maintaining non-negative traffic variables for arbitrary model parameters.
Source: §1 · p. 3The multiplicative Gaussian white-noise assumption is justified using data from leader vehicles accelerating at signalized intersections.
Source: §2 · p. 4
Abstract
from arXiv · showhide
In car-following models, the driver reacts according to his physical and psychological abilities which may change over time. However, most car-following models are deterministic and do not capture the stochastic nature of human perception. It is expected that purely deterministic traffic models may produce unrealistic results due to the stochastic driving behaviors of drivers. This paper is devoted to the development of a distinct car-following model where a stochastic process is adopted to describe the time-varying random acceleration which essentially reflects the random individual perception of driver behavior with respect to the leading vehicle over time. In particular, we apply coupled Langevin equations to model complex human driver behavior. In the proposed model, an extended Cox-Ingersoll-Ross (CIR) stochastic process will be used to describe the stochastic speed of the follower in response to the stimulus of the leader. An important property of the extended CIR process is to enhance the non-negative properties of the stochastic traffic variables (e.g. non-negative speed) for any arbitrary model parameters. Based on stochastic process theories, we derive stochastic linear stability conditions which, for the first time, theoretically capture the effect of the random parameter on traffic instabilities. Our stability results conform to the empirical results that the traffic instability is related to the stochastic nature of traffic flow at the low speed conditions, even when traffic is deemed to be stable from deterministic models.
1. Introduction
Deterministic car-following models leave uncertainty in human perception and driving behavior unrepresented, although stochastic effects are linked to traffic oscillations. The paper proposes a continuous stochastic car-following model and derives stability conditions that capture effects of stochasticity, including instability at low speeds.
1. Introduction
- Existing stochastic traffic-flow approaches include macroscopic methods that add noise to model parameters or represent uncertainty in supply, demand, capacity, or headway.The cited approaches include stochastic fundamental diagrams and stochastic LWR models.Source: §1 · p. 2
- Microscopic car-following research has modeled randomness through desired acceleration, time gaps, and follower free speed, reaction time, or safe distance.The section also notes that changing desired time headway or adding noise to desired acceleration has improved model prediction.Source: §1 · pp. 2–3
- A randomized desired time gap in a discrete IDM has been reported to reproduce synchronized traffic patterns and concave growth of traffic oscillations.The gap varies using a random draw and constants T1 and T2.Source: §1 · p. 3
- Coupled Langevin equations and an extended CIR process represent time-varying driver perception while enhancing non-negative traffic-variable values for arbitrary model parameters.The authors contrast this non-negativity property with models by Laval et al. (2014) and Yuan et al. (2018).Source: §1 · p. 3
- The proposed model's stochastic stability analysis indicates that stochasticity can destabilize traffic at low speeds even when its deterministic counterpart is stable.The introduction says this result conforms to empirical findings and cites Figures 1–3.Source: §1 · p. 3
- The proposed model is calibrated and verified against real-life traffic data, including NGSIM trajectories, with reported good performance.Source: §1 · p. 4
Notation
The notation defines time and vehicle-state variables, followed by parameters governing desired speed, headway, and driver response.
Notation
- t denotes the time instant, while x_n, v_n, Δv_n, and s_n denote vehicle location, speed, relative speed, and bumper-to-bumper gap.The relative speed and gap are defined with respect to the leader, vehicle n−1.Source: Notation · p. 4
- The listed model parameters include desired speed v_0, critical headway s_c, a dimensionless coefficient α, and reaction coefficient β.β is measured in inverse seconds.Source: Notation · p. 4
- σ_0 is identified as the dissipation coefficient, with its unit shown only partially in the passage.The passage gives the unit beginning with √ but does not provide a complete unit.Source: Notation · p. 4
2. Model formulation
The paper formulates car-following as a stochastic process with a deterministic response and a Langevin noise source. It extends CIR dynamics to represent stochastic acceleration while addressing negative speeds and deriving stochastic stability conditions.
2. Model formulation
- The OVM supplies the deterministic car-following response, while the stochastic contribution is intended to extend across a wider class of deterministic models.The authors choose OVM because it suits analytic investigations; they note the stochastic part could also be applied to models such as IDM.Source: §2 · pp. 4–5
- Multiplicative Gaussian white noise represents random acceleration deviations, with a speed-dependent dissipation parameter controlling noise strength.The use of multiplicative noise is justified by collected vehicle data from acceleration at signalized intersections.Source: §2 · pp. 4–5
- The earlier constant-noise model can produce negative speeds at low speed or when σ0 exceeds a predetermined upper bound.Although it can replicate data observations, setting negative speeds to zero in numerical implementation does not remove the model's underlying issue.Source: §2 · p. 5
- The proposed generalization relaxes constant dissipation, uses state-dependent optimal speed, preserves nonnegative speed for arbitrary parameters, and supports generic string stability conditions.These are the stated aims of the extended model.Source: §2 · p. 6
- The extended CIR process makes the noise magnitude depend on speed, so random fluctuations weaken near zero and drift pushes speed toward positive optimal speed.The authors impose Vop = max(0, Vop) in numerical implementation; at equilibrium, CIR speed variance also depends on critical speed vc.Source: §2 · pp. 6–7
3. Linear stochastic stability
The section derives local and string stability conditions for the stochastic car-following model by analyzing small deviations around stationary traffic. The analysis considers multiplicative noise and distinguishes follower recovery from perturbation propagation across followers.
3.1. Local stability condition
- Local analysis uses Lyapunov stability to track whether gap and speed deviations diminish, with leader speed fixed when assessing follower stability.Source: §3.1 · p. 7
- Local stability requires σ0^2 ≤ 8βve; the original OVM (σ0 = 0) is locally stable, while large noise destabilizes it.Source: §3.1 · p. 8
3.2. String stability condition
- String stability analyzes how leader gap and speed perturbations affect followers, deriving a linear condition that accounts for multiplicative white noise.Source: §3.2 · p. 9
- The almost-sure stability condition is stricter than local stability, and mean-square stability implies almost-sure stability.Source: §3.2 · pp. 10–11
- With σ0 = 0, both string stability conditions reduce to the stability condition of the original OVM.Source: §3.2 · p. 11
4. Model performance
The stochastic stability diagrams show that increasing noise can destabilize congested low-speed traffic even when free-flow traffic remains largely unaffected. In the NGSIM calibration, observed speeds fit within the model’s 90% confidence interval and match its predicted mean.
4.1. Stochastic linear stability diagrams
- In congested stable traffic (space headway se ≤20 m), increasing stochasticity destabilizes flow: Figure 1 reports instability below 60 m for σ0 ≥1√ m/s, and Figure 2 below 60 m for σ0 ≥0.5√ m/s.The stable regimes for almost-sure stability are larger than those for mean-square stability.Source: §4.1 · p. 12
- At 18m equilibrium spacing, the OVM is stable at low speed, whereas the SOVM stability condition indicates unstable traffic.The OVM condition is 0.05 > 0; the SOVM condition is 0.1872 < σ0^2 = 1.Source: §4.1 · p. 12, Figure 3 · p. 12
4.2. Model calibration with real data
- Calibration used the congested third NGSIM vehicle group at 800s, with 100 stochastic simulation replications and mean squared error as the performance index.The selected group showed substantial congestion near position 200m.Source: §4.2 · p. 13
- For the second vehicle, observed speeds stayed within the model’s 90% confidence interval and matched the model-output mean; predicted-speed variance was relatively low at low speeds, exemplified at t = 40s.Source: §4.2 · p. 14
5. Concluding remarks
The paper presents a stochastic car-following framework that integrates the OVM to model uncertain human perception and analytically examine traffic instability. Its extended CIR formulation enhances non-negative speed values for arbitrary model parameters; extending the approach to mixed CAV and human-driven traffic is future work.
5. Concluding remarks
- Integrating the OVM into a stochastic car-following framework allows analytical study of human-perception errors and instabilities when the deterministic part is stable.The formulation relaxes constant dissipation and constant optimal-speed assumptions in Laval et al. (2014).Source: §5 · pp. 14–15
- The extended CIR formulation enhances non-negative speed values for arbitrary model parameters.The paper reports that calibration results are consistent with U.S. freeway NGSIM trajectory data.Source: §5 · p. 15
- Future work proposes applying the SOVM to mixed traffic with CAVs and human-driven vehicles to study traffic instability.The authors note that a cited heterogeneous-platoon model considered CAV behavior but ignored stochasticity in human-driven vehicles.Source: §5 · p. 15
Appendix A. Discrete-time stochastic model
The appendix presents a continuous stochastic model in differential form and describes its Euler–Maruyama discretization for simulation. It also reports that changing σ0 has an insignificant effect on speed at low speed, with the change strongly affected by the random term.
Appendix A. Discrete-time stochastic model
- The model is written as du = A(u,t)dt + B(u,t)dW(t) for numerical simulation.The appendix states that the proposed model uses this stochastic differential form.Source: Appendix A · p. 15
- The proposed model is simulated using an explicit Euler–Maruyama scheme.The appendix provides the discretization of the SOVM.Source: Appendix A · p. 16
- At low speed, changing σ0 has an insignificant effect on speed in a time step, and the change is greatly affected by the random term.The passage introduces this as the speed change resulting from a σ0 change of δ.Source: Appendix A · p. 17