Dynamically estimating saturation flow rate at signalized intersections: a data-driven technique

ABSTRACT

Connected Vehicles (CVs) could enhance traffic management systems by providing detailed and real-time information. Theoretically, such information can be exploited for the provision of efficient movement of traffic, especially at intersections identified as the bottlenecks of traffic systems. Aimed at the same purpose, this paper uses information of CVs to estimate the Saturation Flow Rate (SFR), particularly in the transition period during which CVs and conventional vehicles will coexist. To this end, we retain the advantages of data-driven techniques to capture the underlying dynamics of the SFR by considering information of CVs as the only input. In this regard, we correlate the dynamic variations of the SFR to the mutual interactions among the contributing parameters extracted from the limited pieces of CVs’ information using a neural network. Comprehensive simulations under precisely designed settings in VISSIM show a hoped-for SFR estimation accuracy level, which can further augment intelligent intersection controller initiatives.

KEYWORDS:

• Dynamic saturation flow rate
• Connected vehicles
• Mixed traffic
• Neural network
• VISSIM

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Connected vehicles are defined as vehicles that can communicate by any means.

2 The term quality can be interpreted by (1) fitness for the intended purpose, (2) meeting the needed precision, and (3) satisfying simplicity and generalization issues.

3 This concept refers to the power of a model's structure to incorporate contributing elements and to be interpreted by such element.

4 In this case, we are restricted to a single variable equation derived from the average headway.

5 The signal group stands for the group of trajectories that simultaneously receive the right-of-way.

6 A rigorous analysis of the transformation of CVs' data into a suitable form for data mining purposes is left for future research.

7 To calculate the standard deviation, we assume our available data as a sample space.

8 We assume a minimum of 5 vehicles as this threshold.

9 Note that $A_5$ is presented as an auxiliary parameter to boost the model's robustness.

10 By using these approximations, we circumvent the need for a complex cross-validation for regularization.

11 The issue of computation speed is beyond the scope of this investigation. However, the interested readers can refer to (Andayani et al. 2017) for a method of accelerating a backpropagation algorithm.

12 The scripts are available at https://github.com/HossseinMoradi/Project10

13 It should be stressed here that $A_5$ is embedded to compensate for sparse data in case of a low proportion of CVs.

14 For example, in dense networks where intersections are located very close to each other

Additional information

Funding

This work was partially supported by the Research Foundation Flanders (FWO) [grant number 3G051118].