Unit 2

Research Article A Quantitative Method to Delineate the Influence Area of Work Zone Considering Route Choice Inertia and Elastic Demand

Guanfeng Wang ,1 Hongfei Jia ,1 Jingjing Tian ,1 Yu Lin ,2 Ruiyi Wu ,1

Zhendong Liu ,1 and Heyao Gao 1

1College of Transportation, Jilin University, Changchun 130022, China 2Faculty of Maritime and Transportation, Ningbo University, Ningbo 315211, China

Correspondence should be addressed to Hongfei Jia; jiahf@jlu.edu.cn

Received 3 December 2021; Accepted 22 March 2022; Published 20 April 2022

Academic Editor: Lina Kattan

Copyright © 2022 Guanfeng Wang et al. +is 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.

+e work zone on the urban road network will affect the surrounding road traffic. To represent the influence area of the work zone, the concept of a subnetwork is proposed in this paper. Delineating a suitable subnetwork quantitatively is a challenging problem. To address this issue, the node synthesized indexes (NSI) are deployed as a variability measure that captures both the change of link flow and origin-destination (OD) demand. +e inertia-based stochastic user equilibrium with the elastic demand (ISUEED) model is proposed to accurately provide the data of link flow and OD demand for the network with the work zone. Corre- spondingly, the data of the network without a work zone can be obtained by the stochastic user equilibrium (SUE) model. According to the value of NSI, the initial range of the subnetwork is determined. Finally, the connectivity and compactness can be guaranteed by the modified L-shell algorithm. To demonstrate the performance of the method, two case studies and sensitivity analyses are conducted based on the Braess network and the local road network in Changchun, China. +e proposed method is beneficial to reduce the complexity of the traffic model by substituting the entire network with a subnetwork.

1. Introduction

+e long-term presence of a work zone will cause two thorny issues for the traffic network, which are as follows: con- gestion and safety problems [1]. It induces link saturation and causes the accident rate surrounding the work zone to increase noteworthily [2]. Especially, for urban areas with a high road network density, the effects are even more pro- found. Hence, traffic agencies usually implement a series of traffic adjustment measures around the work zone, so that the deterioration of road network performance does not exceed the acceptable threshold.

Generally speaking, the network performance will be better as the application scope of traffic adjustment measures expands. However, it is not desirable in terms of sustainability and economy. +erefore, to achieve a trade-off between the application effect and expenditure of the measures, the sub- network should be delineated appropriately. Unfortunately,

few studies have concentrated on a quantitative method to determine the influence area of the work zone. +e existing literature mainly focuses on judgmental and mathematical methods, which is difficult to meet the requirements of for- mulating specific construction organization scheme [3, 4]. Meanwhile, the quantitative delineation method of the sub- network also puts forward higher requirements on the ac- curacy of data provided by the corresponding traffic models. Although many studies consider route choice inertia and elastic demand separately in traffic assignment models, few researches consider both simultaneously [5–7]. Hence, it is difficult for the model to truly and comprehensively reflect the behavior of traveller, resulting in a reduction in data precision. Furthermore, how to get a connected and compact subnet- work is also a thorny problem. +is study solves the afore- mentioned knowledge gaps by (1) proposing a quantitative delineation method for the subnetwork, (2) building the ISUEED model, and (3) improving the L-shell algorithm [8].

Hindawi Journal of Advanced Transportation Volume 2022, Article ID 2507107, 17 pages https://doi.org/10.1155/2022/2507107

To achieve the above objectives, this study takes nodes as the objects of analysis. A more realistic ISUEED model is developed, which describes the behaviors of travellers in response to an incident by tree structure and exponential elastic demand. +e ISUEED model is applied to obtain the data of link flow and OD demand for the network with work zone. Correspondingly, the data of the network without work zone can be obtained by the SUE model. +e change rates of the two types of data are synthesized into NSI and compared with the threshold value to generate the initial range of the subnetwork. +en the modified L-shell algo- rithm is used to realize the last step of the subnetwork quantitative delineation method. Meanwhile, connectivity and compactness can be guaranteed naturally. Finally, the validity and practicability of the method are verified by the Braess network and the local road network construction project in Changchun, China. Although some parameters are relevant to the case context, the method proposed in this paper is universal.

2. Literature Review

Until now, a large number of pieces of research has con- centrated on mitigating the adverse impact of the work zone on traffic operations [9, 10]. However, the literature mainly focuses on the microsimulation at the link-level and the performance optimization at the network-level.

At the link-level, the researchers realized the optimi- zation of safety, queuing length, time delay, and accident rate by adjusting the length of the work zone, construction time, number of construction teams, and other factors [11–15]. In addition, there is also a way to further divide the work zone into multiple segments to optimize link performance. However, as the number of segments increases, maintenance becomes more expensive. Hence, a compromise between the two needs to be considered [11, 13, 16, 17]. More deeply, to explore the mechanism of action for the work zone under different traffic conditions or vehicle behavior, some re- searchers also conducted simulations for traffic scenes near the work zone, in which the variables include the headway of the vehicle, lane-changing behavior, and following behavior [18–23]. In addition, the duality of new technology is also reflected vividly [24]. For example, the application of the floating car and big data technology provides new oppor- tunities for microsimulation data collection, while the emergence of a moving work zone (a fast construction method using advanced machinery) is a new challenge for relevant research [25].

From the perspective of the network-level, most re- searchers adopt a bilevel programming model. +e upper level aims to minimize the total travel time, traffic delay, pollutant emissions, etc. In contrast, the lower level generally carries out traffic simulations using software, such as VIS- SIM or PARAMICS, and some of them acquire the indicator data through traffic assignment [26–29]. Moreover, most of the studies utilize metaheuristic algorithms to solve the problem of obtaining the optimal work zone scheduling [29, 30]. Among them, several researchers also discuss the factors affecting the schedule of the work zone. Lee

considered that the number of work zones and construction teams would affect the construction order [27]. On the other hand, Rey and Bar-Gera point out the critical role of budget, the number of construction staff, and pieces of equipment [28]. +e length of the construction period and the flexibility of the time window also affect the work zones scheduling [29]. In addition, researchers try distributed algorithms, simplified mathematical distribution models, and other methods to solve the problem of enormous computation burden at the network-level [30].

In urban areas with high road network density, the application of subnetwork has more advantages in re- ducing the complexity of the traffic model and improving the effectiveness of traffic adjustment schemes. However, to our best knowledge, the literature on the subnetwork level is sparse. Firstly, Erath et al. had overcome the challenge of excessive traffic data calculation to assess the vulnerability of the Swiss road network under natural disasters [31]. +e subnetwork division of fixed scope is adopted according to the map, which considers the detour distance constraint of travellers. Secondly, Memarian et al. assessed the impact of lane closures on the surrounding road network during road construction [32]. Subnetwork delineation and a new route choice method are integrated simultaneously to speed up the traffic assignment. Com- pared with the research of Erath et al., this study adopted the mathematical method of linear regression to delineate the subnetwork. It has realized the leap from judgmental approach to quantitative method, however, there are many research gaps.

Previous studies have paid little attention to the sub- network, which also inhibits the development of the rel- evant research of subnetwork delineation methods to a certain extent. In the past, the judgmental approach based on the experience or urban geographic structure was adopted to delineate the subnetwork when a relevant piece of research needs a local part of the network [33, 34]. +ese methods are simple and operable, however, they will bring suboptimal problems. +erefore, some researchers have introduced related methods of the graph theory into the delineation of the subnetwork, most of which are adopted for road user charging cordon design [34]. However, with the expansion of the road network, the shortcomings of this method began to appear, and it became extremely difficult to obtain the initial solution. To overcome it, in recent years, data, such as link density, speed, or travel time, have been used for network clustering to realize subnetwork delineation. Gu and Saberi used the weighted average of density similarity measure and distance similarity measure for graph clustering to achieve the bipartition of the net- work under the premise of ensuring homogeneity, con- nectivity, and compactness [3]. +is method can not only effectively capture the spatial changes of congestion pat- terns in the network but also provide effective insights for the setting of a double layer or even multilayer charging cordon. To sum up, the literature on the subnetwork de- termination method is arid, and the methods currently used have their limitations, respectively. Hence, the pur- pose of this study is to establish a scientific method of

2 Journal of Advanced Transportation

quantitative subnetwork delineation and provide the scope basis for related work zone research.

3. Methodology

+e subnetwork delineation method proposed in this study is displayed in Figure 1. Firstly, the SUE model is used for traffic assignment to obtain the flow and cost of each route between each OD pair when the network is without a work zone. At the same time, the travel demand between each OD pair is classified according to the selected route (multiclass user data). Travellers who choose the same route are known as a class of users. Next, the multiclass user data is treated as the input for the ISUEED model. +en, the data of the link flow and OD demand of the network with a work zone is output as a result of the model. +e NSI is deployed as a variability measure that captures the changes of the network with and without a work zone. Meanwhile, the initial range of the subnetwork can be obtained by comparing NSI with the threshold value. Finally, the L-shell algorithm is used to further process the comparison results to achieve subnet- work delineation. It should be noted that SUE is a mature model and will not be repeated in this article. Please refer to the literature for details [35–37].

To accurately delineate the subnetwork, it is necessary to precisely estimate the data of the link flow and OD demand when there is a work zone on the network. +erefore, the ISUEED model for traffic assignment is proposed, which fully considers the route choice inertial behavior of the traveller and elastic travel demand. +e model takes the multiclass user data obtained from the SUE model as input, represents the route choice inertia behavior of different users using the tree structure, and employs exponential elastic demand to denote demand changes.

3.1. Modeling Route Choice Inertia Behavior. +e change of network topology caused by the existence of the work zone is a new situation for travellers. Research has shown that travellers prefer to stick with past route choice under the circumstances [6]. A tree structure is adopted to describe

this phenomenon in this study, as shown in Figure 2(a). +e reasons are chiefly stated as follows: firstly, it is highly flexible and can correspond to the change of route by the select limb. Secondly, it has favorable adaptability, and the model parameters can be adjusted to adapt to different traveller attributes and sensitivity to cost changes. +irdly, this structure has the advantages of convenience, low data requirements, and accurate calculation results. Specifically, the nests in the upper layer represent the original and currently available set of routes for a specific class of trav- eller. In contrast, the select limbs in the lower layer represent routes. It is essential to point out that the set of original routes contain only one element because their original routes are determined. +e probability calculation processes are expressed as follows:

p m w,i � p

n w,i × p

m|n w,i ∀ w∈W i∈ Iw n∈Nw,i m∈Mw,i,n, (1)

p n w,i �

e −θnμ

n w,i

􏽐n∈Nw,i e

−θnμ n w,i

∀ w∈W i∈ Iw n∈Nw,i, (2)

p m|n w,i �

e −θm c

n,m w,i

􏽐m∈Mw,i,n e −θm c

n,m w,i

∀ w∈W i∈ Iw n∈Nw,i m∈Mw,i,n,

(3)

where pmw,i, p n w,i represent the probability of choosing route

m and nest n for class i travellers between the OD pair w, respectively. pm|nw,i is the conditional probability of choosing route m for class i travellers between OD pair w in the case of choosing nest n. θn, θm are the parameters of the tree structure. μnw,i is the expected minimum travel cost of the nest n for class i travellers between the OD pair w. cn,mw,i is the actual travel cost of route m in the nest n for class i travellers between the OD pair w. W is the set of OD pairs. Iw is the set of original shortest routes (user classes). Nw,i is the set of nests for class i travellers between the OD pair w. Mw,i,n is the set of currently shortest routes for class i travellers between the OD pair w in the case of choosing nest n.

+e specific mechanism of the model to express the route choice inertia behavior is described below. On the network without work zone, travellers choose the route mold. If the route still exists in the shortest route set of the corresponding OD pair on the network with work zone, mold ∈ nnew, this class of travellers tend to maintain their original choices. In this case, the route choice model is the nested logit (NL) model, as shown in Figure 2(b). +e travel cost of the select limb mold in nest nnew is c

mold nnew ⟶∞. In nest nold, it can be

expressed as the travel cost of the route in the network without a work zone multiplied by the inertia coefficient.

c mold nold

� ρcmold, (4)

cmold � 􏽘 l∈L

cl. (5)

+e cost of link can be calculated by the Bureau of Public Roads (BPR) function, which is a strictly monotone function of traffic flow.

OD demand

SUE model

The data of the complete network

ISUEED model

The sub-network

The data of the network with work zone

Modified L-shell algorithm

The change rates of link flow and OD demand

Output

Input

Input

Output Calculate

Output

The node synthesized indexes Input

Calculate Calculate

Figure 1: +e framework of the method.

Journal of Advanced Transportation 3

cl � tf × 1 + α Q

C ( )

β ( ), (6)

where tf is the free-�ow time of link. Q is the tra c �ow on link. C is the capacity of link. α and β are the parameters of the model.

If the route mold no longer exists in the set of the shortest routes because of a topological change, mold ∉ nnew. �e expected minimum travel cost of nold is μnold ⟶∞. �e NL degenerates into a multinomial logit (MNL) model, as shown in Figure 2(c). �e MNL model was ­rst used by Daganzo and She in the route choice of the stochastic tra c assignment. Subsequently, to overcome the inde- pendence of irrelevant alternatives property, the improved versions of the logit model have been developed successively. �e reader is referred to Daganzo and She [35], Prashker and Bekhor [38], and Mai et al. [39]for an in-depth un- derstanding of the logit model in the tra c domain.

3.2. Elastic Demand Function. �e change of network to- pology will lead to the variation of the generalized travel cost of relevant travellers. Travellers may switch their travel mode or cancel their trips. For simplicity, only car travel mode is con- sidered in this study, and the travel mode transfer is not in- volved. As a result, the OD demand will change. Meanwhile, as an important data source of subnetwork delineation, OD de- mand variation cannot be ignored. �erefore, this phenomenon is expressed as an exponential elastic travel demand, as shown in equations (7) and (8) [7]. �e elastic demand function is a strict monotonically decreasing function of the expected minimum travel cost of class i travellers between the OD pair w.

qw,i � Dw,i λw,i( ), (7)

qw,i � q 0 w,ie

φ λw,i

c0w,i − 1 ( )

. (8)

Correspondingly, the inverse function is de­ned as follows:

λw,i � D −1 w,i qw,i( ), (9)

λw,i � c 0 w,i +

c 0 w,i

φ ( )ln

qw,i

q 0 w,i

 , (10)

where qw,i is the travel demand of the class i travellers between the OD pair w. q0w,i is the origin travel demand of the class i travellers between the OD pair w. φ is the elastic coe cient. λw,i is the expected minimum travel cost of the class i travellers between the OD pair w. c0w,i is the travel cost that corresponds to q0w,i, which is acquired using the SUE model.

3.3. Model Formulation. According to equations (1)–(10), the ISUEED model can be described as a variational in- equality (VI) problem. It seeks to minimize the total travel time while ensuring that the tra c �ow complies with the SUE principle (equations (1)–(3)). �e link �ow fn,m∗w,i and travel demand qn∗w,i, q

∗ w,i are de­ned as the optimal solution

of the VI problem:

∑ w∈W

∑ i∈Iw

∑ n∈Nw,i

∑ m∈Mw,i,n

c n,m w,i f

n,m∗ w,i( ) +

1 θm

ln f

n,m∗ w,i

q n∗ w,i

[ ]

· f n,m w,i − f

n,m∗ w,i( ) + ∑

w∈W ∑ i∈Iw

∑ n∈Nw,i

1 θn

ln q

n∗ w,i

q ∗ w,i

q n w,i − q

n∗ w,i( )

− ∑ w∈W

∑ i∈Iw

D −1 w,i q ∗ w,i( ) qw,i − q

∗ w,i( )≥0,

(11)

s.t.

qw,i � ∑ n∈Nw,i

q n w,i∀ w ∈ W i ∈ Iw λw,i( ), (12)

q n w,i � ∑

m∈Mw,i,n

f n,m w,i ∀ w ∈ W i ∈ Iw n ∈Nw,i μ

n w,i( ), (13)

xa � ∑ w∈W

∑ i∈Iw

∑ n∈Nw,i

∑ m∈Mw,i,n

f n,m w,i δ

n,m w,i,a∀ a ∈A, (14)

f n,m w,i ≥0∀ w ∈ W i ∈ Iw n ∈ Nw,i m ∈Mw,i,n τ

n,m w,i( ), (15)

qw,i≥0∀ w ∈ W i ∈ Iw ψw,i( ), (16)

where fn,mw,i is the �ow of route m in the nest n for class i travellers between the OD pair w. qnw,i is the travel demand of nest n for class i travellers between the OD pair w. xa is the �ow of link a. δ

n,m w,i,a represents whether

Route choice

nold

mold mold mkm1 m2

nnew

......

(a)

nold

mold mkm1 m2

nnew

Route choice

......

(b)

mkm1 m2

nnew

Route choice

......

(c)

Figure 2: �e tree structure of route choice.

4 Journal of Advanced Transportation

the link a is on the route m in the nest n for class i travellers between the OD pair w. If so, δn,mw,i,a � 1, other- wise δn,mw,i,a � 0.

+e constraints given by equations (12) and (13) denote the flow conservation equations. +e constraint equation (14) represents the relationship function between the route flow and the link flow. Equations (15) and (16) denote the non-negative constraints. +e corresponding dual variables are in parentheses.

Proposition 1. 3e VI problem defined by equations (11–16) is equivalent to equations (1)–(10).

Proof. Equation (11) is expressed as z1, z2, and z3, re- spectively, as follows:

z1 � 􏽘 w∈W

􏽘 i∈Iw

􏽘 n∈Nw,i

􏽘 m∈Mw,i,n

c n,m w,i f

n,m∗ w,i􏼐 􏼑 +

1 θm

ln f

n,m∗ w,i

q n∗ w,i

􏼢 􏼣

· f n,m w,i − f

n,m∗ w,i􏼐 􏼑,

z2 � 􏽘 w∈W

􏽘 i∈Iw

􏽘 n∈Nw,i

1 θn

ln q

n∗ w,i

q ∗ w,i

q n w,i − q

n∗ w,i􏼐 􏼑,

z3 � 􏽘 w∈W

􏽘 i∈Iw

D −1 w,i q ∗ w,i􏼐 􏼑 qw,i − q

∗ w,i􏼐 􏼑.

(17)

+e Lagrange function is constructed by the Lagrange multiplier method.

L � z1 + z2 + z3 + λw,i qw,i − 􏽘 n∈Nw,i

q n w,i

⎛⎝ ⎞⎠

+ μnw,i q n w,i − 􏽘

m∈Mw,i,n

f n,m w,i

⎛⎝ ⎞⎠.

(18)

+en Karush–Kuhn–Tucker (KKT) conditions can be expressed as follows:

f n,m w,i c

n,m w,i +

1 θm

ln f

n,m w,i

q n w,i

− μnw,i􏼠 􏼡 � 0,

c n,m w,i +

1 θm

ln f

n,m w,i

q n w,i

− μnw,i≥0,

f n,m w,i ≥0,

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(19)

q n w,i

1 θn

ln q

n w,i

qw,i − λw,i + μ

n w,i􏼠 􏼡 � 0,

1 θn

ln q

n w,i

qw,i − λw,i + μ

n w,i≥0,

q n w,i≥0,

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(20)

qw,i −D −1 w,i qw,i􏼐 􏼑 + λw,i􏼐 􏼑 � 0,

−D −1 w,i qw,i􏼐 􏼑 + λw,i≥0,

qw,i≥0,

⎧⎪⎪⎪⎨

⎪⎪⎪⎩

(21)

Based on equation (19), equations (22)–(24) can be obtained.

c n,m w,i +

1 θm

ln f

n,m w,i

q n w,i

− μnw,i � 0, (22)

ln f

n,m w,i

q n w,i

� θm μ n w,i − c

n,m w,i􏼐 􏼑, (23)

f n,m w,i

q n w,i

� e θmμ

n w,i × e

− θm c n,m w,i . (24)

Add the traffic flow of m routes.

1 � eθmμ n w,i 􏽘

m∈Mw,i,n

e −θm c

n,m w,i .

(25)

+en, we can acquire equation (26), which is equivalent to equation (3).

f n,m w,i

q n w,i

� p m|n w,i

� e

−θm c n,m w,i

􏽐m∈Mw,i,n e

−θm c n,m w,i

.

(26)

Similarly, we can obtain equations (27) and (28) from equation (20), where equation (28) is equivalent to equation (2).

λw,i � − 1 θn

ln 􏽘 n∈Nw,i

e −θnμ

n w,i . (27)

q n w,i

qw,i � p

n w,i

� e

− θnμ n w,i

􏽐n∈Nw,i e −θnμ

n w,i

.

(28)

Lastly, based on equation (21), (29) can be obtained, which is equivalent to equation (8).

qw,i � Dw,i λw,i􏼐 􏼑. (29) □

Proposition 2. 3e VI problem has a unique optimal solution.

Proof. According to equation (11), the Hessian matrix corresponding to z1, z2 , and z3 are positive definite. +us, the model is strictly convex. Besides, the constraints are all linear. +erefore, this problem is convex programming. As shown in equations (4)–(6), the travel cost of the link is

Journal of Advanced Transportation 5

continuous with respect to xa, and the route cost is con- tinuous with the route flow. +en, any local optimal solution must be a global optimal solution, and it is unique. In sum, the VI problem has a unique optimal solution. □

3.4. Solution Algorithm for the ISUEED Model. +e ISUEED model essentially is a multiclass user problem, which can be solved by the diagonalized method of the successive average (DMSA) algorithm [40–42]. It is worth noting that according to the definition of the multiclass user in this study, only one OD pair is involved per user class. +erefore, the travel demand is loaded according to the route choice model, and the equilibrium state is achieved simultaneously in the process of single-class user assignment. As a result, it eliminates the traffic adjustment process within the per class of user. Even though the number of user classes is enormous, the total running time is still acceptable. To meet the needs of this study, the link flow X and OD demand set WE are taken as the output results. Compared to previous literature, the algorithm has been slightly adjusted. Algorithm 1 gives the corresponding solving steps specifically.

3.5. Delineating the Subnetwork. +is study takes nodes as the analysis object. +e change rates of the link flow and

OD demand with and without a work zone are used to generate the NSI. +en, we compare the NSI with the threshold value to determine whether a node is in the influence area. Finally, the subnetwork is delineated by the affected nodes, which are selected by the L-shell algorithm.

+e change rate of the link flow for node u is related to the incoming flow and outgoing flow, which are calculated, respectively, as follows:

ΔFinu � 􏽐∈Mu

f m u −􏽐m′∈Mu′

f m’ u

􏼌􏼌􏼌􏼌􏼌

􏼌􏼌􏼌􏼌􏼌

􏽐m∈Mu f

m u

,

ΔFoutu � 􏽐m∈Mu

f m u −􏽐m′∈Mu′

f m’ u

􏼌􏼌􏼌􏼌􏼌

􏼌􏼌􏼌􏼌􏼌

􏽐m∈Mu f ′m u 􏽐

,

(30)

where ΔFinu ,ΔF out u are the change rates of the incoming

(outgoing) flow at node u. fmu , f ’m u are the flows that enter

(leave) node u through link m in the network without work zone. fm’u , f

’m’ u are the flows that enter (leave) node u through

link m’ in the network with work zone. Mu, Mu′ are the sets of links connected by node u when the network is with (without) a work zone.

+e average value of the two is calculated as the change rate of flow for node u and normalized.

Input: the network G (N, A) with work zone, Iw(∀w ∈W), set of travel demand for each class travellers between OD pair w Qw(∀w ∈W), convergence accuracy ξ. Output: X, WE

(1) By the k-shortest paths method to generate the set of shortest routes Iw′(∀w ∈ W). (2) Initialize: assign Qw(∀w ∈ W) according to the SUE model. Acquire the traffic flow (3) of the route f1

w,i′ and the traffic flow of the link x1a. e ⟶∞. For n � 1.

(4) while e>ξ (5) for w in W (6) Update the link travel time. (7) Update the route travel time. (8) Update the expected minimum travel cost μw between each OD pair w. (9) for i in Iw(∀w ∈W) (10) Updated the OD qnw,i of such travellers according to equation (8). (11) if i ∈ Iw′(∀w ∈W) (12) Correct route travel cost according to equation (6). (13) Load qnw,i to each route according to the NL shown in equations. (14) (1)–(3) (15) else (16) Load qnw,i to each route according to the MNL. (17) end if (18) end for (19) end for (20) Update the link flow xna. (21) Acquire additional traffic flow of the route gn

w,i′ .

(22) Update the route traffic flow fn+1 w,i′

� fn w,i′

+ χn(g n w,i′

− fn w,i′

), χn � 1/n.

(23) Computational convergence precision e � ��������������������� 􏽐w∈W􏽐i′∈Iw′

(fn+1 w,i′

− fn w,i′

) 2

􏽱 /􏽐w∈W􏽐i′∈Iw′f

n w,i′

. (24) For n � n + 1 (25) end while (26) Calculate qnw � 􏽐

i∈Iw qnw,i(∀w ∈W).

(27) Store each xna and each q n w to X and WE, respectively.

(28) Output X and WE.

ALGORITHM 1: DMSA.

6 Journal of Advanced Transportation

ΔFaveu � ΔFinu +ΔF

out u

2 ,

ΔFu � ΔFaveu −ΔF

ave min

ΔFavemax −ΔF ave min

,

(31)

whereΔFaveu is the average value of �ow change rate at node u.ΔFavemax, ΔF

ave min are the maximum and minimum values of

the average �ow change rate.ΔFu is the normalized value of ΔFaveu .

�e change rate of OD demand for node u is tightly associated with the generating demand and attracting de- mand, which are calculated, respectively, as follows:

ΔQgenu � ∑n∈Nd

n u −∑n∈Nd

n’ u

∣∣∣∣∣

∣∣∣∣∣

∑n∈Nd n u

,

ΔQattu � ∑n∈Nd

’n u −∑n∈Nd

’n’ u

∣∣∣∣∣

∣∣∣∣∣

∑n∈Nd ’n u

,

(32)

where ΔQgenu , ΔQattu are the change rates of the generation (attraction) demand of node u. dnu, d

’n u are the demands

generation (attraction) of node u to node n when the net- work is without a work zone. dn’u , d

’n’ u are the demands

generation (attraction) of node u to node n when the net- work is with a work zone.

Similarly, the change rate of OD demand for node u is as follows:

ΔQaveu � ΔQgenu +ΔQ

att u

2 ,

ΔQu � ΔQaveu −ΔQ

ave min

ΔQavemax −ΔQ ave min

,

(33)

whereΔQaveu is the average value of demand change rate at node u. ΔQavemax, ΔQ

ave min are the maximum and minimum

values of the average demand change rate. ΔQu is the normalized value of ΔQaveu .

Based on equations (31) and (33), we de­ne NSI for each node u in the network as a weighted average ofΔQu andΔFu.

Su � αΔQu +(1 − α)ΔFu, (34)

where Su is the value of NSI for node u. α is the weight coe cient.

�e NSI is compared with the threshold to determine whether node u is within the area of in�uence, if so, put it into the set E.

state � in Su≥Ssta out Su⟨Ssta

{ ∀u ∈ N, (35)

where Ssta is the threshold value of NSI, which can be ad- justed from 0 to 1 according to various needs to generate di¡erent subnetworks.

3.6. Modi�ed L-Shell Algorithm. �e L-shell algorithm is a method of community division in the complex network. �is

27 28 29 13 14 30 31

25 26 12 6 5 15 32

24 11 7 1 2 4 16

23 22 10 8 3 17

21 20 19 9 18 35 34

33 Level 1

Level 2

(a)

27 28 29 13 14 30 31

25 26 12 6 5 15 32

24 11 7 1 2 4 16

23 22 10 8 3 17

21 20 19 9 18 35 34

33 Sub-network

(b)

27 28 29 13 14 30 31

25 26 12 6 5 15 32

24 11 7 1 2 4 16

23 22 10 8 3 17

21 20 19 9 18 35 34

33

Sub-network

(c)

Figure 3: Examples of related concepts in the L-shell algorithm.

Journal of Advanced Transportation 7

section only brie�y explains the part involved in this study. �e speci­c details and algorithm processes are shown in the literature [8]. In Figure 3(a), nodes 1 and 2 are vital nodes. Nodes 6, 7, and 8 are the adjacent nodes of node 1 in level 1. Nodes 12 and 13 are the level 1 adjacent nodes of node 6 and the level 2 adjacent nodes of node 1. According to the above process, it can be seen that the nodes 3–8 in Figure 3(a) are the level 1 adjacent nodes of vital nodes 1 and 2, denoted as the set l1. �e nodes 9–18 are all the level 2 nodes of vital nodes 1 and 2, denoted as the set l2.

�e distribution of NSI is not necessarily a centralized radial but may be sporadic and disconnected. To ensure the connectivity and compactness of the subnetwork, a modi­ed L-shell algorithm is proposed to perform secondary ­ltering on nodes. �e algorithm takes set E as input. It outputs the set F composed of all nodes within the subnetwork. �e speci­c steps are shown in Algorithm 2. �e central idea is to use the ­rst “while” loop to basically determine the sub- network, and then use the second “while” loop to check and ­ll gaps. In the ­rst loop, it goes through over the elements in set En and places it in set F if its level 1 adjacent nodes already have at least one node in set F. Repeat the process until the size of set En does not change, which represents that all nodes in set E that can connect to a network have been selected. �is process can be described as the vital nodes constantly adsorb the eligible nodes in set En layer-by-layer around it. �e second loop is equivalent to a tolerance device, which is mainly to ensure the integrity of the sub- network. Compared with the ­rst loop, it changes both in the value of level and the judgment condition of putting the node into the set F. By traversing all the nodes that are not in

set F, if the level 1 or level 2 adjacent nodes of one node are already in set F, put it in set F. For example, as shown in Figure 3(b), node 4 is not taken into set F after the above procedure. However, it is also going to be put in F in this loop. However, if the number of nodes exceeds the threshold, as depicted in Figure 3(c), nodes 1–7 are con- sidered to be outside the subnetwork.

4. Case Study

We conduct two case studies to demonstrate the properties of the proposed method. Firstly, the Braess network is used to illustrate the e¡ectiveness of the ISUEED model. Sub- sequently, we apply the Guilin Road network to test the feasibility and adaptability with respect to a larger scale network.

4.1. �e Braess Network. �e ISUEED model is the crucial component of the method proposed in this paper. �erefore, we select a small example to manifest the improvement of

Input: the set of nodes E filtered by a threshold, the set of nodes N in the network. Output: the set of nodes F which compose of sub-network.

(1) Initialize: put the vital nodes into F, E1 �∅, E2 � E, n � 2. (2) while En≠En− 1 (3) for e in En (4) Taking node e as the starting point, search all level 1 adjacent nodes of node (5) e according to the L-shell algorithm and store to Ne. (6) Determine whether there is at least one member of Ne that is already (7) included in F, if so, put e into F, otherwise, put it into En+1. (8) end for (9) For n � n + 1 (10) end while (11) Calculate Y � N − E + En. (12) Initialize: Y1 �∅, Y2 � Y, n � 2. (13) while Yn≠Yn− 1 (14) for y in Yn (15) Taking node y as the starting point, search all level 1 adjacent nodes and (16) level 2 adjacent nodes of y according to the L-shell algorithm and store (17) to Ny1 , Ny2, respectively. (18) Determine whether the members of Ny1 or Ny2 are all included in F, if (19) so, put y into F, otherwise, put it into Yn+1. (20) end for (21) For n � n + 1 (22) end while (23) Output F.

ALGORITHM 2: Modi­ed L-shell algorithm.

O

B

A

D

1 2

3

4

5

6

Figure 4: �e Braess network.

8 Journal of Advanced Transportation

Table 1: �e properties of links.

Link Length (m)

Capacity (veh/h）

Free-�ow time(sec)

α β

1 400 60 20 0.15 4 2 400 60 20 0.15 4 3 400 60 20 0.15 4 4 400 60 20 0.15 4 5 400 60 20 0.15 4 6 600 60 30 0.15 4

Table 2: �e �ow of links and routes.

Before interruption

After interruption

Original (veh/h)

SUE (veh/h)

SUEED (veh/h)

ISUEED (veh/h)

Link 1 32.20 64.53 62.81 62.41 Link 2 32.20 35.47 34.53 34.93 Link 3 32.20 35.47 34.53 34.93 Link 4 — 29.06 28.28 27.48 Link 5 32.20 64.53 62.81 62.41 Link 6 35.60 — — — Route 1 32.20 35.47 34.53 34.93 Route 2 32.20 35.47 34.53 34.93 Route 3 — 29.06 28.28 27.48 Route 4 35.60 — — — OD demand 100 veh 100 veh 97.34 veh 97.34 veh

Work zone

Study area

Changchun,China

Ziyou Rd

Jiefang Rd

R enm

in A ve

X inm

in A ve

N

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

Work zone

Figure 5: Case study work zone in the Guilin Road.

Journal of Advanced Transportation 9

the ISUEED model. As shown in Figure 4, the network includes 4 nodes, 6 links, 1 OD pair, and 4 routes. �e OD demand is 100 veh and the 4 routes are as follows: route 1: links 1-2; route 2: links 3–5; route 3: links 1-4-5; route 4: link 6. In the original network, the three valid routes are 1, 2, and 4. We assume that link 6 is interrupted. �e valid routes in the network are routes 1, 2, and 3. Detailed information about the network is shown in Table 1.

Table 2 depicts the �ow of each route and link before and after link 6 interruption. To illustrate the improvement of the model, we specially compare the results of the SUE, SUEED, and ISUEED models, which can present the in�uence of elasticity and inertia on tra c �ow. First of all, we can observe that because of the interruption of link 6, the travel demand decreases from 100 to 97.34, with the expected travel cost between node O and node D increasing. Secondly, as travellers have a preference for routes 1 and 2, inertia leads to a slight decrease in the number of travellers choosing route 3, which is con­rmed by the results of the SUEED and ISUEED models. To sum up, these ­ndings demonstrate that the ISUEED model proposed in this study has distinguished performance. By introducing elasticity and inertia, it can

portray the travel behavior of travellers more realistically and obtain tra c assignment results closer to reality.

4.2. �e Guilin Road Network. To explore the properties of the proposed method in subnetwork delineation, a case study is carried out based on the road maintenance project of the Guilin Road business district in Changchun, China. Since the method in this study takes nodes as the analysis object, the real road network is abstracted into a grid net- work for the convenience of expression. As shown in Fig- ure 5, it has 136 nodes and 482 links. �ere is a work zone between node 65 and node 66, which is 200 meters in length. It involves the construction of pipeline, cable, and pavement projects. �e planned construction period is two months. During this period, the lanes are closed in both directions, and all vehicles are required to bypass the surrounding roads. �e road is a branch with one lane in each direction, and its peak hour two-way tra c volume is 1566 veh/h. �e link �ow of the network with and without work zone are obtained from ­eld tra c surveys. �e corresponding travel demand data is derived from the OD estimation by TransCAD software.

Th reshold=0.8

Th reshold=0.5

Th reshold=0.2

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

Figure 6: Subnetwork with di¡erent thresholds.

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

A ctual scope of influence

Figure 7: Subnetwork of ­eld survey.

10 Journal of Advanced Transportation

Table 3: Statistics on the change rates of OD demand and link �ow.

OD demand Link �ow Change rate (%) Number Change rate (%) Number [−1.05, −0.90] 7573 [−100, −90] 2 (−0.9, −0.75] 2567 (−90, −80] 0 (−0.75, −0.60] 2045 (−80, −70] 2 (−0.60, −0.45] 1247 (−70, −60] 0 (−0.45, −0.30] 799 (−60, −50] 0 (−0.30, −0.15] 529 (−50, −40] 5 (−0.15, 0] 882 (−40, −30] 7 (0, 0.15] 457 (−30, −20] 13 (0.15, 0.30] 448 (−20, −10] 21 (0.30, 0.45] 435 (−10, 0] 63 (0.45, 0.60] 371 (0, 10] 57 (0.60, 0.75] 313 (10, 20] 75 (0.75, 0.90] 246 (20, 30] 76 (0.90, 1.05] 154 (30, 40] 47 (1.05, 1.30] 110 (40, 50] 41 (1.30, 1.45] 74 (50, 60] 31 (1.45, 1.60] 38 (60, 70] 17 (1.60, 1.75] 32 (70, 80] 8 (1.75, 1.90] 19 (80, 90] 5 (1.90, 2.05] 10 (90, 100] 6 (2.05, 2.30] 5 ≥100 8 (2.30, 2.45] 4 (2.45, 2.60] 1 (2.60, 2.75] 1

1 8

16 24 32 40 48 56 64 72 80 88 96

104

1 8 16 24 32 40 48 56 64 72 80 88 96 10 4

112

11 2

120

12 0

128

12 8

136

13 6

0.020

0.015

0.010

0.05

0

–0.05

–0.010

Figure 8: �e change rate of OD matrix.

1 12 23 34 45 56 67 78 89 10 0

11 1

12 2

13 3

14 4

15 5

16 6

17 7

18 8

19 9

21 0

22 1

23 2

24 3

25 4

26 5

27 6

28 7

29 8

30 9

32 0

33 1

34 2

35 3

36 4

37 5

38 6

39 7

40 8

41 9

43 0

44 1

45 2

46 3

47 4

Link ID

-1.5 -1

-0.5 0

0.5 1

1.5 2

2.5

R at

e of

c ha

ng e

Figure 9: �e change rate of link saturation.

Journal of Advanced Transportation 11

0.10

0.15

0.20

0.25

0.30

0.35

N SI

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.1 Inertial coefficient

Figure 10: Box plot of NSI at di¡erent inertia.

Table 4: Subnetwork constitution under di¡erent parameters.

Parameter Number of nodes Number of links ANSI

Inertia coe cient

0.1 5 8 0.2 0.4 20 52 0.25 0.7 39 116 0.24 1 56 174 0.22

Elastic coe cient

−0.15 63 208 0.23 −0.09 49 158 0.27 −0.06 30 90 0.25 −0.01 7 12 0.22

Capacity 0 49 158 0.35 0.5 22 62 0.27 0.8 10 22 0.22

Scenarios

1 35 110 0.25 2 23 68 0.26 3 45 148 0.23

Coexisting 99 342 0.21

OD distribution Upper 58 186 0.23 Right 46 122 0.22

Surrounding 70 256 0.23

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

N SI

–0.01–0.03–0.06–0.09–0.12–0.15–0.18 Elastic coefficient

Figure 11: Box plot of NSI at di¡erent elasticity.

12 Journal of Advanced Transportation

4.3. Delineation Results. �e subnetwork of work zone in the road network is determined by the delineation method pro- posed in this study. With respect to the increasing threshold, Figure 6 depicts an obviously smaller subnetwork. When the threshold value is 0.8, it means that the nodes with extremely drastic changes would be classi­ed into the subnetwork. Hence, it can be seen that the subnetwork is only limited to the ad- jacent links of the work zone. To verify the feasibility and accuracy of the method, this study obtains the data of the link �ow before and after work zone existence from ­eld investi- gation. Taking the 20% change rate of link saturation as the criterion, the actual area of in�uence is determined as shown in Figure 7. It is basically consistent with the subnetwork de- lineated by the proposed method with the threshold value of 0.2. On the other hand, with the decrease of threshold value, the spatial variation of the subnetwork is captured, which re�ects the spatial propagation trend of network disturbance to a great extent. It is bene­cial to achieve the network congestion prediction and targeted intervention. Furthermore, the sub- network under di¡erent requirements can be realized by adjusting the threshold value. For example, when the tra c management agency has a higher requirement on the per- formance of the road network, the threshold can be lowered to make the subnetwork larger, and vice versa.

Although the change rates of link �ow and OD demand are only the subsidiary products of the calculation of the NSI, their appearance provides new perspectives for the subse- quent formulation of targeted tra c adjustment strategies. Figure 8 comprehensively summarizes the spatial distribu- tion of the change rates of OD demand among nodes, which appears as regional aggregation and cyclical change. As displayed in Table 3, more than 90% of OD pairs slightly decrease by about 1%, which corresponds to the blue part in Figure 8. �is phenomenon also re�ects that the elastic demand will keep it stable by itself, and the changes only occur in a narrow area. Public transportation and other modes should be prepared in advance to undertake this part of transferred OD demand and avoid greater disturbance to the tra c network. �erefore, the reasonable cooperation of various transportation modes should be an important part of the transportation organization scheme of work zone. On the contrary, there are only a few parts in red, which are OD demands that vary dramatically. It is worth noting that the white part in Figure 8 indicates that OD demand is 0, that is to say, there is no travel between the corresponding two zones. From a spatial point of view, there is an obvious negative correlation between the change rate of link satu- ration and its distance from the work zone (link ID � 224

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

Elastic coeffi cient=-0.01

Elastic coeffi cient=-0.06

Elastic coeffi cient=-0.09

Elastic coeffi cient=-0.15

Figure 13: Subnetwork under di¡erent elasticity.

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

Inertia coeffi cient=0.1

Inertia coeffi cient=0.4

Inertia coeffi cient=0.7

Inertia coeffi cient=1

Figure 12: Subnetwork under di¡erent inertia.

Journal of Advanced Transportation 13

and 227), as shown in Figure 9. �e adjacent links of the work zone have changed even more dramatically. According to the statistical results in Table 3, the change rate of link saturation is mainly between −100% and 150%, and only one link is close to 200%, presenting a normal distribution with 0 as the expectation on the whole. Tra c managers should focus on these links because slight improvements may signi­cantly boost overall network performance.

4.4.SensitivityAnalysis. Table 4 shows the statistical results of the three indicators with the subnetwork under dif- ferent parameter scenarios, which are the number of nodes, the number of links, and the average value of NSI (ANSI). It is interesting to note that ANSI has no obvious trend of change. �ese results might attribute to the second loop of the modi­ed L-shell algorithm. As the selection criteria is relaxed, there may be an inde­nite number of nodes whose NSI is inferior to the threshold to be delineated into the subnetwork. Hence, ANSI is an unpredictable value, which may be related to the OD demand distribution and road network topology. �e

speci­c action mechanism of each parameter is displayed in Figures 10–17, respectively.

Firstly, we explore the e¡ect of inertial route choice behavior and elastic demand on sub-network delineation. Scenarios with di¡erent inertia coe cients and elastic co- e cients are discussed. As shown in Figure 10, with the inertial coe cient increasing from 0.1 to 1 at step size 0.1, the maximum, minimum, and average values of NSI present an upward trend at ­rst and then increase slowly. Meanwhile, Figure 12 illustrates the subnetworks with inertial coe - cients of 0.1, 0.4, 0.7, and 1, respectively, which present a signi­cant tendency to expand. �e main reason for this is that travellers are more inclined to choose the fresh route as the coe cient goes up, and the disturbance to the road network is drastic, and vice versa. When the inertia coef- ­cient is close to 0, there will be an extreme scene, where almost all travellers travel on the original route, unless the route no longer exists because of a change of network to- pology. Hence, the network changes are only limited to a narrow scope. To sum up the aforementioned conclusions, it reminds that tra c managers can guide the travellers to choose the original route as much as possible through media publicity, economic leverage, and road tra c restrictions.

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

W ork zone 1

W ork zone 2

W ork zone 3

C oexisting

Figure 15: Subnetwork under di¡erent scenarios.

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

C apacity=0.8

C apacity=0.5

C apacity=0

Figure 14: Subnetwork under di¡erent disturbances.

14 Journal of Advanced Transportation

After all, it can e¡ectively reduce the disturbance to the network.

On the contrary, from Figure 11, it can be seen that as the elastic coe cient gradually increases from −0.18 to −0.01, and the box plots of NSI show an obvious trend of decline. �e subnetwork (elastic coe cient � −0.15, −0.09, −0.06, and −0.01) will diminish at the same time, as shown in Figure 13. It is well-known that OD demand is tightly connected with the generalized travel costs of related routes. Travellers with greater �exibility are more likely to shift their travel mode or even cancel their trips when they are faced with the situation of work zone existing in the network, which has deep in�uence on the OD demand. Meanwhile, it suggests that maybe changing the cost of private car or other alternative travel modes is a feasible way to achieve minimal degradation of network performance.

To explore the in�uence of di¡erent disturbance degrees on subnetwork delineation, three scenarios, including single-lane interruption, one-way interruption, and two-way interruption, are proposed. �ey assume that the corresponding capacities are 80%, 50%, and 0% of the original capacity, respectively. As shown in Figure 14, with the decrease of construction intensity, its subnetwork will shrink signi­cantly. In view of the above

reasons, it is not recommended to completely cut o¡ con- struction, especially if the construction period is long. Inversely, the gap of the impact caused by di¡erent methods may be marginal when the construction period shortens. On the other hand, it also emphasizes the importance of designing indi- vidualized construction schemes according to the character- istics of di¡erent projects.

In some cases, multiple work zones will exist at the same time, which will have a superimposed impact on the road network. To investigate the impact of this situation on the mechanism of subnetwork delineation, we assumed that there are three work zones located between node 29 and node 30, node 55 and node 56, and node 85 and node 86. Figure 15 depicts the area of in�uence when three work zones exist alone and coexist. �e di¡erent locations of work zone are equivalent to the di¡erent space coverages of subnetworks, and they overlap with each other. However, it should be noted that when multiple work zones coexist, their subnetworks are not simply superimposed but redistribute the overall network tra c �ow. �erefore, the joint organization and optimiza- tion of multiple work zones is the preferred way. It also provides powerful insights for the design of the construction schedule with multiple work zones.

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

U pper area

R ight area

Surrounding area

Figure 16: �e situation of area division.

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30 31 32 33

34 35 36 37 38 39

42 43 44 45 46 47 48 49 50

51 52 53 54 55 56

57 58

61 62 63 64 65 66

70 71 72 73 74 75 76 77 78

67 68 69

81 82 83 84 85 86 87 88 89

79 80

96 97 98 99 100 101 102 103 104

90 91 92 93 94 95

105 106 107 108 109 110 111 112

113 114 115 116 117 118 119 120

121 122 123 124 125 126 127

129 130 131 132 133 134 135 136

40 41

59 60

128

O riginal

U pper

R ight

Surrounding

Figure 17: Subnetwork under di¡erent OD distributions.

Journal of Advanced Transportation 15

Furthermore, the effect of OD distribution on sub- network is also discussed briefly. We assume the following three OD distributions: firstly, the OD demand of the upper area of the work zone increases twice as its origin (upper). Secondly, the OD demand of the right area of the work zone increases twice as its origin (right). +irdly, the OD de- mand in the surrounding area of work zone remains the same, however, the OD demand outside the surrounding area increases twice as its origin (surrounding). +e specific area division and corresponding subnetwork are shown in Figures 16 and 17. It can be observed that with the change of OD demand, the subnetwork has obvious deviation. Specifically, the subnetwork will extend to the area with a larger OD demand and display distinct asymmetry to the work zone. +is phenomenon also presents that it is necessary to fully investigate OD distribution before de- signing traffic adjustment scheme, which will make the scheme more efficient.

5. Conclusion

+is paper realizes the quantitative delineation of the sub- network, which provides a scientific scope basis for the evaluation of traffic impact and the formulation of traffic measures. +is method takes the NSI as a variability measure to capture both the changes of link flow and OD demand. +e ISUEED model fully reflects the response of the network to the incident by route choice behavior and localized de- mand change. +erefore, it can accurately provide the data of link flow and OD demand for the network with work zone. Correspondingly, the SUE model is applied to obtain the data of the network without a work zone. +en, by com- paring NSI with the threshold value, the initial range of the subnetwork can be determined. Finally, the modified L-shell algorithm is employed to ensure the connectivity and compactness of the subnetwork.

+e results not only show the feasibility and practica- bility of the proposed subnetwork delineation method but also demonstrate its robustness to parameters and scenario changes. Hence, managers can adjust parameters to obtain different subnetworks that meet the personalized require- ments of various projects. At the same time, it also gives the changes of link flow and OD demand in the spatial di- mension, which is of great help to the development of traffic management plan. On the other hand, the subnetwork with different interruption modes, the coexistence of multiple work zones, and OD distributions are also conducted. +e conclusions can be used to clarify the direction of the construction scheme formulation.

+e current research can be widened in the following aspects: firstly, day-to-day traffic assignment models can be applied in the framework to investigate the cumulative loss during the construction period. Secondly, since it is ex- pensive for data collection, advanced data acquisition technologies should be introduced. Finally, the popularity of connected autonomous vehicles is an irreversible trend, which will influence the network performance deeply. +us, there exists the necessity to explore the subnetwork delin- eation under mixed traffic environment.

Data Availability

All data and program files included in this study are available from the corresponding author upon request.

Conflicts of Interest

+e authors declare that they have no conflicts of interest.

Acknowledgments

+is work was supported by Science and Technology De- velopment Project of Jilin Province under Grant 20190303124SF.

References

[1] J. Weng, Q. Meng, and T. F. Fwa, “Vehicle headway distri- bution in work zones,” Transportmetrica: Transportation Science, vol. 10, no. 4, pp. 285–303, 2014.

[2] E. Rista, T. Barrette, R. Hamzeie, P. Savolainen, and T. J. Gates, “Work zone safety performance: comparison of alternative traffic control strategies,” Transportation Research Record: Journal of the Transportation Research Board, vol. 2617, no. 1, pp. 87–93, 2017.

[3] Z. Gu and M. Saberi, “A bi-partitioning approach to con- gestion pattern recognition in a congested monocentric city,” Transportation Research Part C: Emerging Technologies, vol. 109, pp. 305–320, 2019.

[4] A. Memarian, S. P. Mattingly, J. M. Rosenberger, J. C. Williams, S. A. Ardekani, and H. Hashemi, “Modeling framework to identify an affected area for developing traffic management strategies,” Journal of Transportation Engi- neering, Part A: Systems, vol. 144, no. 10, Article ID 04018059, 2018.

[5] C. Xie and Z. Liu, “On the stochastic network equilibrium with heterogeneous choice inertia,” Transportation Research Part B: Methodological, vol. 66, pp. 90–109, 2014.

[6] J. Zhang and H. Yang, “Modeling route choice inertia in network equilibrium with heterogeneous prevailing choice sets,” Transportation Research Part C: Emerging Technologies, vol. 57, pp. 42–54, 2015.

[7] T. Zhao and Y. Zhang, “Transportation infrastructure res- toration optimization considering mobility and accessibility in resilience measures,” Transportation Research Part C: Emerging Technologies, vol. 117, Article ID 102700, 2020.

[8] J. P. Bagrow and E. M. Bollt, “Local method for detecting communities,” Physical Review E, vol. 72, no. 4, Article ID 046108, 2005.

[9] J. Kianfar and S. Abdoli, “Deterministic and stochastic ca- pacity in work zones: findings from a long-term work zone,” Journal of Transportation Engineering, Part A: Systems, vol. 147, no. 1, Article ID 04020141, 2021.

[10] W. Zhu, Z. Li, J. Ash, Y. Wang, and X. Hua, “Capacity modeling and control optimization for a two-lane highway lane-closure work zone,” Journal of Transportation Engi- neering, Part A: Systems, vol. 143, no. 12, Article ID 04017059, 2017.

[11] D. Cao, J. Wu, J. Wu, B. Kulcsár, and X. Qu, “A platoon regulation algorithm to improve the traffic performance of highway work zones,” Computer-Aided Civil and Infra- structure Engineering, vol. 36, no. 7, pp. 941–956, 2021.

[12] Y. Chung and W. W. Recker, “A methodological approach for estimating temporal and spatial extent of delays caused by

16 Journal of Advanced Transportation

freeway accidents,” IEEE Transactions on Intelligent Trans- portation Systems, vol. 13, no. 3, pp. 1454–1461, 2012.

[13] Q. Meng and J. Weng, “Optimal subwork zone length and project start time for short-term daytime work zones from the contractor’s perspective,” Transportation Research Part C: Emerging Technologies, vol. 29, pp. 72–83, 2013.

[14] G. Pesti and R. E. Brydia, “Work zone impact assessment methods and applications,” Transportation Research Record: Journal of the Transportation Research Board, vol. 2617, no. 1, pp. 52–59, 2017.

[15] L. Zhao, S. I. Chien, and B. Du, “Optimizing work-zone schedule with floating car data considering traffic diversion and managed lanes,” Journal of Transportation Engineering, Part A: Systems, vol. 145, no. 1, Article ID 04018076, 2019.

[16] H. Sharma, C. McIntyre, Z. Gao, and T.-H. Nguyen, “De- veloping a traffic closure integrated linear schedule for highway rehabilitation projects,” Journal of Construction Engineering and Management, vol. 135, no. 3, pp. 146–155, 2009.

[17] C. Lu, J. Dong, A. Sharma, T. Huang, and S. Knickerbocker, “Predicting freeway work zone capacity distribution based on logistic speed-density models,” Journal of Advanced Trans- portation, vol. 201815 pages, 2018.

[18] H. Abdulsattar, A. Mostafizi, and H. Wang, “Surrogate safety assessment of work zone rear-end collisions in a connected vehicle environment: agent-based modeling framework,” Journal of Transportation Engineering, Part A: Systems, vol. 144, no. 8, Article ID 04018038, 2018.

[19] X. Gan, J. Weng, W. Li, and M. Han, “Spatial-temporal varying coefficient model for lane-changing behavior in work zone merging areas,” Journal of Transportation Safety & Se- curity, pp. 1–24, 2020.

[20] N. Raju, S. Arkatkar, and G. Joshi, “Effect of construction work zone on traffic stream parameters using vehicular tra- jectory data under mixed traffic conditions,” Journal of Transportation Engineering, Part A: Systems, vol. 146, no. 6, Article ID 05020002, 2020.

[21] J. Weng, G. Du, D. Li, and Y. Yu, “Time-varying mixed logit model for vehicle merging behavior in work zone merging areas,” Accident Analysis & Prevention, vol. 117, pp. 328–339, 2018.

[22] J. Weng, X. Gan, and G. Du, “Random coefficient models for work zone headway distribution,” Journal of Transportation Engineering, Part A: Systems, vol. 145, no. 10, Article ID 04019042, 2019.

[23] K. Zhang and M. Hassan, “Identifying the factors contributing to injury severity in work zone rear-end crashes,” Journal of Advanced Transportation, vol. 20199 pages, 2019.

[24] B. Du, S. Chien, J. Lee, and L. Spasovic, “Predicting freeway work zone delays and costs with a hybrid machine-learning model,” Journal of Advanced Transportation, vol. 20178 pages, 2017.

[25] X. Gan, J. Weng, and J. Zhang, “Evaluation of travel delay and accident risk at moving work zones,” Journal of Trans- portation Safety & Security, vol. 13, no. 6, pp. 622–641, 2021.

[26] A. Alvanchi, M. Rahimi, M. Mousavi, and H. Alikhani, “Construction schedule, an influential factor on air pollution in urban infrastructure projects,” Journal of Cleaner Pro- duction, vol. 255, Article ID 120222, 2020.

[27] H.-Y. Lee, “Optimizing schedule for improving the traffic impact of work zone on roads,” Automation in Construction, vol. 18, no. 8, pp. 1034–1044, 2009.

[28] D. Rey and H. Bar-Gera, “Long-term scheduling for road network disaster recovery,” International Journal of Disaster Risk Reduction, vol. 42, Article ID 101353, 2020.

[29] H. Zheng, E. Nava, and Y.-C. Chiu, “Measuring networkwide traffic delay in schedule optimization for work-zone planning in urban networks,” IEEE Transactions on Intelligent Trans- portation Systems, vol. 15, no. 6, pp. 2595–2604, 2014.

[30] W. Ma, R. Long Cheu, and D.-H. Lee, “Scheduling of lane closures using genetic algorithms with traffic assignments and distributed simulations,” Journal of Transportation Engi- neering, vol. 130, no. 3, pp. 322–329, 2004.

[31] A. Erath, J. Birdsall, K. W. Axhausen, and R. Hajdin, “Vul- nerability assessment methodology for Swiss road network,” Transportation Research Record: Journal of the Transportation Research Board, vol. 2137, no. 1, pp. 118–126, 2009.

[32] A. Memarian, J. M. Rosenberger, S. P. Mattingly, J. C. Williams, and H. Hashemi, “An optimization-based traffic diversion model during construction closures,” Com- puter-Aided Civil and Infrastructure Engineering, vol. 34, no. 12, pp. 1087–1099, 2019.

[33] A. D. May, R. Liu, S. P. Shepherd, and A. Sumalee, “+e impact of cordon design on the performance of road pricing schemes,” Transport Policy, vol. 9, no. 3, pp. 209–220, 2002.

[34] A. Sumalee, “Optimal road user charging cordon design: a heuristic optimization approach,” Computer-Aided Civil and Infrastructure Engineering, vol. 19, no. 5, pp. 377–392, 2004.

[35] C. F. Daganzo and Y. Sheffi, “On stochastic models of traffic assignment,” Transportation Science, vol. 11, no. 3, pp. 253–274, 1977.

[36] W. H. K. Lam, Z. Y. Gao, K. S. Chan, and H. Yang, “A stochastic user equilibrium assignment model for congested transit networks,” Transportation Research Part B: Method- ological, vol. 33, no. 5, pp. 351–368, 1999.

[37] M. J. Maher, “Stochastic user equilibrium assignment with elastic demand,” Traffic Engineering and Control, vol. 42, no. 5, pp. 163–167, 2001.

[38] J. N. Prashker and S. Bekhor, “Route choice models used in the stochastic user equilibrium problem: a review,” Transport Reviews, vol. 24, no. 4, pp. 437–463, 2004.

[39] T. Mai, M. Fosgerau, and E. Frejinger, “A nested recursive logit model for route choice analysis,” Transportation Re- search Part B: Methodological, vol. 75, pp. 100–112, 2015.

[40] Y. Lin, H. Jia, B. Zou et al., “Multiobjective environmentally sustainable optimal design of dedicated connected autono- mous vehicle lanes,” Sustainability, vol. 13, no. 6, 3454 pages, 2021.

[41] M. Fujita, S. Yamada, and S. Murakami, “Time coefficient estimation for hourly origin-destination demand from ob- served link flow based on semidynamic traffic assignment,” Journal of Advanced Transportation, vol. 2017, pp. 1–14, 2017.

[42] L. de Grange, M. Marechal, and F. González, “A traffic as- signment model based on link densities,” Journal of Advanced Transportation, vol. 2019, pp. 1–20, 2019.

Journal of Advanced Transportation 17

Copyright of Journal of Advanced Transportation is the property of Hindawi Limited and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.