Journal of Advanced Transportation

Volume 2017 (2017), Article ID 9243039, 8 pages

https://doi.org/10.1155/2017/9243039

## A Trial-and-Error Method with Autonomous Vehicle-to-Infrastructure Traffic Counts for Cordon-Based Congestion Pricing

^{1}Jiangsu Key Laboratory of Urban ITS, Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies, School of Transportation, Southeast University, Jiangsu, China^{2}Department of Logistics & Maritime Studies, The Hong Kong Polytechnic University, Hung Hom, Hong Kong

Correspondence should be addressed to Zhibin Li

Received 19 April 2017; Accepted 31 May 2017; Published 2 July 2017

Academic Editor: Xiaopeng Li

Copyright © 2017 Zhiyuan Liu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

This study proposes a practical trial-and-error method to solve the optimal toll design problem of cordon-based pricing, where only the traffic counts autonomously collected on the entry links of the pricing cordon are needed. With the fast development and adoption of vehicle-to-infrastructure (V2I) facilities, it is very convenient to autonomously collect these data. Two practical properties of the cordon-based pricing are further considered in this article: the toll charge on each entry of one pricing cordon is identical; the total inbound flow to one cordon should be restricted in order to maintain the traffic conditions within the cordon area. Then, the stochastic user equilibrium (SUE) with asymmetric link travel time functions is used to assess each feasible toll pattern. Based on a variational inequality (VI) model for the optimal toll pattern, this study proposes a theoretically convergent trial-and-error method for the addressed problem, where only traffic counts data are needed. Finally, the proposed method is verified based on a numerical network example.

#### 1. Introduction

Traffic congestion has become one of the key problems for the operation of urban transport systems. Traffic congestion not only is a daily source of frustration for many motorists but also results in environmental issues and social expenses. Congestion causes longer travel times, increased travel time variability, air and noise pollution, and lower productivity in the work force [1–5]. Road pricing has been well recognized as an important tool for the land transport authority to manage the traffic demand, shifting transit ridership [6–8], and eventually ameliorate the traffic congestion; see Small and Verhoef [9], Lawphongpanich et al. [10], and Cheng et al. [11] among many others. A prosperous literature is observed for the theoretical studies of urban congestion pricing. Many studies have focused on determining the optimal toll fares. Two economics concepts, first-best pricing (no restriction on the pricing locations) and second-best pricing (only part of the network is tolled), have been widely discussed in the context of urban congestion pricing. Despite their theoretical soundness, the achievements of many studies on toll determination are not suitable for practical implementations, mainly because they require the accurate data on many network attributes, including travel demand function and link travel time functions as well as network users’ value-of-time [12], and it is difficult to obtain these data for the whole transport network. This article thus accounts for the practical implementation of congestion pricing by solving the toll determination problem while obviating these data.

Since the practical implementation of urban congestion pricing is an important focus of this study, we then investigate some existing congestion pricing practices in Singapore since 1975 [13], in London since 2003 [14], and in Stockholm since 2007 [15]. These three well-known practices all adopt cordon-based schemes, where London uses an area licensing charge and the other two cities take entry-based charge. The entry-based charge is known to be more equitable and efficient than area licensing charge [16]. Therefore, this article also takes cordon-based scheme with entry-based charge as a target.

Two practical properties of the cordon-based pricing are further considered in this article. Firstly, when cordon-based pricing is implemented in practice, the network authority is concerned more about the traffic conditions within the cordon area; for instance, when the congestion pricing was first established in Singapore in 1975, Singapore Land Transport Authority’s original target of implementing this scheme is to reduce the total traffic volume to the cordon area by 25% to 35% [13]. This article thus further takes into consideration this point; namely, the total inbound traffic volume of one cordon should be restricted by a predetermined threshold. Secondly, the toll charge on each entry to one cordon should be identical, in view of the convenience for recognition by the drivers and for administration by the authority; for example, toll charge on each link is the same at the Orchard Cordon and Bugis-Marina Centre of Electronic Road Pricing (ERP) system in Singapore.

To sum up, this article aims to propose a method for the toll determination of such a cordon-based pricing, where massive data is not needed. Traffic flow data is of crucial value for the studies of urban transport system, for example, the studies on traffic signaling [17, 18] as well as traffic safety studies [19]. With the fast development of intelligent transportation systems and the Internet of things, it is very convenient to collect the number of vehicles in any particular link. Hence, the proposed trial-and-error method only needs the autonomous traffic counts data on the tolled links, which can be easily gathered based on the vehicle-to-infrastructure (V2I) communications.

#### 2. Literature Review

The concept of obviating network attributes data and instead using traffic survey data to adjust the toll fares until achieving the optimal tolls is termed as “trial-and-error” method in the literature. Based on the seminal works by Downs [20] and Vickrey [21], Li [22] first proposed a trial-and-error method for marginal cost pricing based on a simple two-link example, where the Origin-Destination (OD) demand function is avoided. This work by Li has been well recognized in the literature, and only recently Wang and Yang [23] pointed out that under some extreme conditions (when the demand function is too flat) this solution method could not converge, and a bisection-based trial-and-error method was then developed to remedy this problem. Such a concept of trial-and-error is also of considerable importance for other sorts of transportation networks [24, 25].

Yang et al. [26] proposed a trial-and-error method for the first-best pricing in a general transport network assuming deterministic user equilibrium (DUE) circumstances, which was then extended to the logit-based stochastic user equilibrium (SUE) case by Zhao and Kockelman [27]. As an extension of Yang et al. [26], Han and Yang [28] have proposed a more efficient trial-and-error method for the first-best pricing using advanced step sizes. In the context of second-best pricing, Yang et al. [29] then proposed a convergent trial-and-error mechanism for the optimal toll pattern, which is recently extended by Ye et al. [30] by considering the day-to-day dynamics of network flows.

The aforementioned trial-and-error methods obviate the origin-destination (OD) demand function, yet they still rely on the data of link travel time functions and drivers’ value-of-time, which are not all readily available in practice. Based on the concept of traffic-restraint pricing (the objective is to restrict traffic volumes on some links to a desirable target level), Meng et al. [31] have proposed a trial-and-error method that only requires the traffic counts on certain links. This traffic count data could be efficiently and accurately obtained in practice from the toll booths or detection loops. Thus, it is an engineering-based method suitable for practical use. Yang et al. [32] then extended this work by considering elastic demand and asymmetric link flow interactions. In the context of general SUE, Meng and Liu [12] built a trial-and-error method based on the gradient projection algorithm, where the convergence can be guaranteed under mild conditions. Zhou et al. [33] developed a new trial-and-error method for the first-best congestion pricing solution with restriction to link capacity constraints. Recently, Xu et al. [34] have developed a trial-and-error method for the traffic-restraint congestion pricing scheme in a network with day-to-day flow dynamics.

As to the cordon-based pricing, these existing methods in the literature, however, are not valid to use, especially considering the two practical properties of cordon-based pricing addressed in this article, namely, (a) the total inbound flow to one cordon should be restricted and (b) toll charge on each link to one cordon should be identical. Considering the wide practicality of cordon-based pricing in reality, an extension of the previous trial-and-error methods for cordon-based pricing is of considerable significance.

Assuming the users’ route choice behavior follows the General SUE principle; this article proposed a trial-and-error method based on the variational inequality (VI) model built by Liu et al. [35] for the side-constrained SUE problem. The total inbound traffic flow to each cordon is limited by a predetermined threshold value, which is taken as a side constraint to the SUE traffic assignment problem. It is shown that the optimal toll pattern equals the optimal Lagrangian multiplier of the side constraints.

The remainder of this article is organized as follows. The next section describes the connection between the optimal toll pattern and side constraints to the cordons. Then, a convergent trial-and-error method for the optimal toll pattern is developed, which is numerically verified by a network example. Finally, the conclusions and future works are presented.

#### 3. Problem Description and Mathematical Model

##### 3.1. Assumptions and Notations

Let denote a strongly connected network; the attributes of this network are represented by the following notations::set of nodes:set of links; is the cardinality of set :set of OD pairs; is the cardinality of :travel demand between OD pair :column vector of all the OD travel demands; :set of all the paths between OD pair , and is the cardinality of : if path between OD pair traverses link ; otherwise:link/path incidence matrix associated with OD pair ; namely, :link/path incidence matrix for the entire network; :traffic flow on path between OD pair :column vector of traffic flows on all the paths between OD pair ; namely, :column vector of traffic flow on all the paths in the network; that is, :OD pair/path incidence matrix, , where equals 1 if path and 0, otherwise:traffic flow on link :column vector of all the link traffic flows, :asymmetric link travel time function on link , and it is a nonnegative, monotonically increasing, and continuously differentiable function of the link flow vector :travel time on path , and :column vector of travel times on all the paths connecting OD pair .

According to the flow conservation law, path flows, link flows, and OD travel demands should fulfill the following equations:Equations (1) define the feasible set for path flows and link flows, denoted by and , respectively.

Assume there are totally cordons in the network, and all the entry links to cordon , , are charging with the same toll fare denoted by . The toll charges of all the cordons are grouped into vector . Let denote the set of all the entry links to cordon , , and all the tolled links are then grouped into set . Denoted by , the toll charge on link equals if and if link is not an entry link to any cordon. Different toll pattern will affect the drivers’ route choice and thus give rise to different equilibrium flows. Let denote the equilibrium link flow on link in terms of the following generalized link travel time functions:where is the drivers’ value-of-time. It should be noted that this value-of-time is actually not needed when the trial-and-error method proposed below is used in practice.

As mentioned before, the optimal toll pattern is one that can restrict the inbound flow to each cordon to a predetermined threshold , namely, To further reflect the equity issue, the toll charge should be zero if the total inbound flows are strictly less than the threshold value; that is,In addition, the toll charge of each cordon should be nonnegative,Herein, (3), (4), and (5) define the mathematical conditions for the optimal toll pattern . The objective of this article then becomes to find a trial-and-error method for such an optimal toll pattern based on only traffic count data, which can be easily gathered based on the V2I facilities on the road side.

##### 3.2. Mathematical Model

We first discuss the mathematical model for the traffic assignment problem, namely, how to get the equilibrium link flows in terms of a given toll pattern . In view of its better representativeness, the general SUE is adopted to depict the drivers route choice behavior. Hence, with given toll charge pattern , the drivers make their route choice decisions based on the perceived path travel timewhere the random variable is assumed to cover the perception error of all the drivers across the whole population on the travel time of path . In the literature, is usually assumed to follow Gumbel distribution (Logit-based SUE) or normal distribution (probit-based SUE); see Liu et al. [36, 37].

For the general SUE problem with asymmetric link travel time functions, the following fixed point model proposed by Daganzo [38] can be used to formulate the traffic assignment problemwhere is the path choice probability equal to the probability that is perceived as the shortest path among all the paths in set . Equation (7) can be solved by the convergent Method of Successive Average [39].

We proceed to build a mathematical model for the optimal toll pattern that can fulfill (3), (4), and (5). In fact, since the feasible set of toll pattern is the whole positive orthant, the nonlinear complementarity conditions (3), (4), and (5) are equivalent to the following variational inequality (VI) problem [40]:where is the feasible set of and the VI function is defined as

Hence, solving the VI model gives the optimal toll pattern that can satisfy the mathematical conditions (3), (4), and (5). The monotonicity of this type of VI model in the context of cordon-based pricing has been rigorously proven by Liu et al. [35], which guarantees the convergence of many projection-based solution algorithms. It is shown in the following section that a trial-and-error mechanism can be built based on the solution algorithm for the VI model (8).

#### 4. Trial-and-Error Method

Many projection-type methods can be observed in the literature for solving the VI models proposed for transport problems, for example, the works by Nagurney [41], Meng and Liu [39], and Zhou et al. [42]. In this section, we adopt the self-adaptive predictor-corrector (PC) method proposed by He and Liao [43], which is an extension of the PC method with advanced self-adaptive step sizes. We see that this method can converge linearly, which is efficient for practical implementations. The self-adaptive PC method takes the calculation of projection operations of a toll charge pattern as a subroutine. In view of the fact that the feasible set of toll vector here is the nonnegative orthant, calculating the projection operation is very efficient and convenient; for example, for a vector , its projection to the feasible set of the toll patterns, denoted by , equals

Accordingly, model (8) can be efficiently solved using the self-adaptive PC method. The main steps of the self-adaptive PC method are then summarized as follows.

*Step ** 0 *(initialization). Choose an initial vector , three constants , , and initial step size . Let the number of iterations .

*Step ** 1 *(prediction process).

*Step ** 1.1.* Implement the toll pattern in the network and then observe the traffic flows on the entry links to each cordon, which are denoted by , and then compute .

*Step ** 1.2*. Find auxiliary vector by the projection:*Step 1.3*. Consequently, levy the toll pattern in the network and then count the corresponding traffic flows on the entry links, denoted by , which is used to calculate .

*Step ** 1.4*. Calculate ratio byIf , go to Step 2; otherwise reduce the step size according toand go to Step .

*Step ** 2 *(correction process). Based on , , and , calculate a step size for correction and then get an updated vector .

*Step 2.1*. Calculate another step size as per the formula where*Step 2.2*. Update the vector by this projection:*Step 3. *Slightly enlarge step size according to the following scheme:*Step 4 *(stop check). Ifwhere is a predetermined positive tolerance, then stop; otherwise, let and go to Step .

The output of PC method gives the optimal toll pattern , and it can be seen from the solution procedure that only traffic count data autonomously collected on entry links in Steps 1.1 and 1.3 are needed in practice. Thus, this method can avoid gathering the data of OD demand, link travel time functions, or users’ VOT that are extremely difficult to be accurately collected.

#### 5. Numerical Example

A numerical example is employed in this section to validate the proposed model and effectiveness of the self-adaptive PC method. As shown in Figure 1, this example contains 7 nodes, 11 links, and one pricing cordon. The cordon area is defined by its vertex nodes 1, 4, 5, and 7. The three entry links 5, 6, and 7 to this cordon are charging. Denoted by , the optimal toll charges on the entry links are identical.