Research Article  Open Access
Lei Zhao, Hongzhi Guan, Junze Zhu, Yunfeng Wei, "A ReliabilityBased Network Equilibrium Model with Adaptive RiskAverse Travelers", Journal of Advanced Transportation, vol. 2018, Article ID 5294185, 13 pages, 2018. https://doi.org/10.1155/2018/5294185
A ReliabilityBased Network Equilibrium Model with Adaptive RiskAverse Travelers
Abstract
In this paper, route freeflow travel time is taken as the lower bound of route travel time to examine its impacts on budget time and reliability for degradable transportation networks. A truncated probability density distribution with respect to route travel time is proposed and the corresponding travel time budget (TTB) model is derived. The budget time and reliability are compared between TTB models with and without truncated travel time distribution. Under truncated travel time distribution, the riskaverse levels of travelers are adaptive, which are affected by the characteristics of the used routes besides the confidence level of travelers. Then, a TTBbased stochastic user equilibrium (SUE) is developed to model travelersâ€™ route choice behavior. Moreover, its equivalent variational inequality (VI) problem is formulated and a routebased algorithm is used to solve the proposed model. Numerical results indicate that route travel time boundary produces a great influence on decision cost and route choice behavior of travelers.
1. Introduction
Most studies indicate that travel time reliability (TTR) is an essential travel decision cost of travelers [1â€“3]. In particular, AbdelAty et al. (1977) pointed out that TTR was one of the major elements influencing commuters' route choice considerations [4]. However, the classical user equilibrium (UE) principle assumes that travelers are riskneutral and their decision behavior only depends on the mean travel time in deterministic transportation networks. Obviously, it leaves the travel time uncertainty out of consideration and neglects the risk preferences of travelers (i.e., risk aversion and risk seeking). Meanwhile, empirical studies demonstrated that majority of travelers are riskneutral and they would like to take an additional payment to avoid the congestion and travel risk [5, 6].
In transportation systems, uncertainty is unavoidable, which mainly derives from roadway capacity variation and travel demand fluctuation. Both supply and demand variations bring about travel time uncertainty. In order to model travelersâ€™ route choice behavior under uncertainty, a variety of traffic equilibrium models incorporating TTR are developed [7â€“11]. Typically, Lo et al. (2006) proposed the within budget time reliability (WBTR) model [12]. The notion of travel time budget (TTB) in WBTR model is given to represent the decision cost of travelers. The TTB refers to the sum of the mean travel time and a safety margin to arrive on time. The standard deviation of route travel time is used to measure the risk of route choice. However, the WBTR model disregards the effects of unreliable side of route travel time on route choice. By incorporating simultaneously the reliable and unreliable sides, the meanexcess traffic equilibrium (METE) model was developed by Chen and Zhou (2010) [13]. In the METE model, travelers are assumed to use such a route to make TTR ensured and unreliable impacts can be minimized. Subsequently, Chen et al. (2011) further introduced the perception errors of travelers to the METE model and proposed the stochastic meanexcess traffic equilibrium (SMETE) model [14]. In this case, the perceived travel time distribution is derived to model travelers' route choice behavior rather than the actual one. Different from the previous studies, Watling (2006) considered travelers' schedule delay and proposed a late arrive penalized user equilibrium (LAPUE) model [15]. Recently, the multiobjective equilibrium problems have been intensively investigated [16â€“18]. In this regard, Wang et al. (2014) provided a biobjective user equilibrium (BUE) condition, which is the state that no traveler can improve either his/her expected travel time or risk or both without worsening the other by unilaterally changing routes [17].
It can be found that most of aforementioned traffic equilibrium models with the TTR are based on the degradable transportation networks. As mentioned above, travel demand fluctuation would also lead to travel time uncertainty. Shao et al. (2006) developed the UE model with demand uncertainty [19]. Then, the model was further extended by considering travelersâ€™ perception errors and heterogeneity, and the reliability stochastic user equilibrium (RSUE) model was developed [20]. Lam et al. (2008) assumed the uncertainties mainly due to adverse weather condition and proposed a traffic equilibrium model with doubly uncertainties in supply and demand sides [21]. Zhou and Chen (2008) compared and analyzed the equilibrium results of three UE models with stochastic travel demand, using a simple network to illustrate the differences, and examined how these models address the travel risk [22]. Siu and Lo (2008) formulated a multiclass equilibrium model, in which the demand composes of infrequent travelers and commuters [23]. Sun and Gao (2012) assumed the probability distributions of travel demand and link capacities are unknown and developed the robust traffic equilibrium model [24]. In addition, considered uncertainties in demand and supply are also extended to the network design problems (NDPs) to develop the robust optimization models [25, 26].
In the above models, route travel time are commonly assumed to follow continuous normal distribution or lognormal distribution. However, these distributions imply that the lower boundary of route travel time approaches negative infinity or zero. To the best of our knowledge, route travel time is bounded and there is at least a lower boundary, namely, freeflow travel time. Yan et al. (2015) studied the effects of link speed limits on travelersâ€™ route choice and the network performance [27]. This study indicated that the mean and variance of total travel time can be reduced by imposing speed limits, but the total TTB of a network increases. Xu et al. (2017) showed clearly that the minimum travel time equals the link length divided by the speed limit value after imposing a speed limit scheme [28]. They indicated that a speed limit scheme on uncertain road networks would affect the network flow reallocation. In the study, the minimum travel time is derived from speed limits. Nevertheless, the route travel time without a speed limit should also have a lower bound, and the travel time uncertain profile would also change.
This paper aims to examine the impacts of route travel time boundary on travel cost and route choice of travelers. To this end, we firstly propose the probability density function (PDF) with the route freeflow travel time as a truncation of original distribution. Subsequently, the TTB model with truncated travel time distribution is derived and compared with the original TTB model in terms of budget time and TTR. Then, the TTBbased SUE models are developed to model the impacts of route travel time boundary on travelersâ€™ route choice.
The remainder of this paper is organized as follows. Section 2 presents the effect of route travel time boundary on PDF. Section 3 develops the TTB model with truncated travel time distribution under uncertainty. Section 4 gives the TTBbased SUE model. Then, we use numerical examples to illustrate the effect of route travel time boundary on network equilibrium in Section 5. Finally, Section 6 summarizes some concluding remarks.
2. Effect of Route Travel Time Boundary on Uncertainty
In this section, the route freeflow time is considered as the lower bound of route travel time, and the effects of route travel time boundary on uncertainty are examined by proposing a truncated PDF of route travel time. The relationships and differences between TTB models with and without route travel time boundary are derived.
Let us consider a stochastic network , which is composed of nodes and directed links with capacity variations. Let be the freeflow time, the traffic flow, and the capacity of link , respectively. Then, link travel time can be determined by the bureau of public roads (BPR) link performance function, i.e.,where and are, respectively, the deterministic parameters of the BPR link performance function. Assume that link capacity follows an uniform distribution, , where is the design capacity associated with link and denotes the capacity degradable coefficient, . Further, we suppose that link capacity distributions are independent, and thus route travel time is normally distributed ,â€‰â€‰ based on the central limit theorem. The mean and standard deviation of route travel time are, respectively,where and are, respectively, the mean and standard deviation of link travel time. is a binary variable, which equals 1 when link is on route , and 0 otherwise. is a set of all routes between OD pair , and is a set of all OD pairs in the transportation network.
Combining the derivation of the mean and standard deviation of link travel time in [12], (2) and (3) can be rewritten as follows.
2.1. Truncated Route Travel Time Distribution
This paper assumes that the route freeflow time is a determinate constant associated with specific route in the transportation network. Travel time uncertainty is only caused by the link capacity degradation. As derived above, route travel time is normally distributed under such case. A normal distribution implies that the lower bound or minimum route travel time is close to negative infinity. However, the minimum route travel time must be a positive constant, namely, the route freeflow travel time. In other words, the freeflow travel time is the lower bound of route travel time, and route travel time cannot be less than the freeflow travel time. More specifically, the probability of route travel time being less than the freeflow travel time should be zero. Hence, we develop the PDF with the freeflow travel time as a truncation of original distribution without the lower bound, i.e.,where and are the PDF and CDF of the original normal distribution associated with route travel time . is the route freeflow travel time and . and are the mean and standard deviation of route travel time, respectively.
For the ease of elaboration, a singleroute network with the mean of 20 and the standard deviation of 5 is used. Assume the freeflow travel time is 15. The PDFs of route travel time with and without lower bound are shown in Figure 1. The xaxis shows the change in route travel time. The blue solid line is the PDF curve without the lower bound. The green dotted line represents the PDF curve with the freeflow travel time as a truncation of the original distribution. From this figure, one can see that the original distribution is truncated and becomes taller due to the appearance of fixed denominator. Moreover, the probability equals zero when route travel time with the lower bound is less than freeflow travel time. Obviously, the new PDF still fulfills the conservation property of a valid PDF according to (6). In addition, rather than truncated continuous distributions, other modeling ways can also be used to model the impacts of the lower bound of route travel time. Xu et al. (2018) elaborated the reality and feasibility of truncated travel time distribution and its mathematical tractability [28].
2.2. Mean and Standard Deviation Derivations
Below, we derive the moments of route travel time with the freeflow travel time as the lower bound. The truncated PDF in (6) can be rewritten aswhere is the CDF of the standard normal distribution. The mean and standard deviation of truncated travel time distribution are derived using the method of moment estimation as follows.
The mean of truncated travel time distribution can be derived, i.e.,
Let , we have and
Then, substituting (4) into (3), we have the following.
Obviously, it can be observed that from (5). It implies that the mean of truncated route travel time moves right and becomes larger compared to the original one.
Similarly, the secondary moment of truncated travel time distribution can be derived, i.e.,
Then, the variance and standard deviation of truncated travel time distribution are, respectively,
3. TTBBased Route Choice Model with Truncated Travel Time Distribution
In order to model travelersâ€™ route choice behavior under risk, different decision criterions have been proposed, such as TTB, METE, and PMETE. Xu et al. (2014) provided a detailed review on existing traffic equilibrium models under uncertainty [29]. This section restates the concepts of TTB and TTR and illustrates the effects of the route travel time boundary on them.
3.1. TTB without Route Travel Time Boundary
Definition 1. The TTB on route between OD pair associated with a predefined confidence level is defined as the sum of the expected travel time of travelers and a buffer time, the latter of which is the product of the standard deviation and travelersâ€™ riskaversion level , i.e.,
Definition 2. The TTR refers to the probability that travelers can arrive on time within the TTB , i.e.,Thus, (15) can also be alternatively written as follows.Mathematically, the riskaversion level can be expressed aswhere is the inverse function of the CDF of standard normal distribution. Obviously, travelersâ€™ riskaversion level is only related to their expected ontime arrival probability . In other words, the scale of for users with the same confidence level within one OD pair is identical.
3.2. TTB with Route Travel Time Boundary
The route freeflow time is viewed as a truncation of the original normal distribution, the resultant PDF is significantly different. In this case, the TTB and TTR with travel time boundary need to be derived and discussed.
For ease of the mathematical derivation, we assume that the expected travel time and risk have no changes under the truncated travel time distribution, and the TTB with route travel time boundary is aswhere and are the TTB and riskaversion level under truncated travel time distribution, respectively. is no longer the constant which is only related to travelersâ€™ confidence level and it is also affected by the characteristics used routes (i.e., route travel time distribution and route freeflow travel time). More specifically, travelersâ€™ riskaversion level is contextbased and adaptive, which would be proved and derived in the following.
According to Definition 2, the TTR under truncated distribution can be expressed as
Obviously, (19) can also be written as the conditional probability, i.e.,where is the freeflow travel time on route between OD pair . Mathematically, (20) can be alternatively written as
According to (21), we have the following.
Then, we can derive
Let , we havewhere denotes the inverse function of the CDF of normal distribution.
3.3. Effect of Truncated Travel Time Distribution on TTB and TTR
In this subsection, some theoretical analysis on the TTB and TTR changes is given when considering route travel time boundary, and we illustrate the effects of truncated travel time distribution on TTB and TTR.
Proposition 3. For route between OD pair , the truncated travel time distribution makes the corresponding TTR reduce under the same budget time compared to the original distribution.
Proof. Assume and denote the within budget time reliability with and without travel time boundary, respectively, and the relationship between and can be expressed aswhere is used in the above inequality. From (25), is smaller than , and the proof of Proposition 3 is completed. In addition, Proposition 3 also implies that ignoring route travel time boundary would result in overestimating the TTR.
Proposition 4. For route between OD pair , the truncated travel time distribution provides a longer TTB under the same confidence level compared to the original distribution.
Proof. Assume and are the TTBs with and without travel time boundary, the WBTR under truncated travel time distribution can be expressed as follows.Then, we transform (26) and have the following.Similarly, the TTR under nontruncated travel time distribution can be derived, i.e.,Meanwhile, (28) can be written as According to (27) and (29), we can obtain the difference of budget time between two types of distribution, i.e.,Because the CDF of normal distribution is monotonically increasing, . Proposition 4 is proved.
Proposition 5. Travelers adopt a selfadaptive riskaversion level under truncated travel time distribution, which is not only related to travelersâ€™ confidence level, but also affected by the used route.
Proof. Combining (18) and (22), the TTR under truncated travel time distribution can be derieved, i.e., By rearranging terms of LHS in (31), and Then, can be derived as follows.Because is associated with and , the riskaversion level is not only related to travelersâ€™ confidence level, but also affected by the used route. In addition, the difference of riskaversion levels between different distributions can be expressed as follows.Obviously, , and is monotonically increasing function. So we have . It implies that route travel time boundary has a great effect on travelersâ€™ risk attitudes.
3.4. An Illustrative Example
An example network with two parallel routes connecting one origin and one destination is used to present the difference between the TTB model with and without truncated travel time distribution. In this example, the route travel time is assumed to follow a normal distribution . The means and standard deviations of two route travel time are assumed to be , , , . The freeflow travel times are 15 min and 10 min, respectively. All users are assumed to be homogeneous with the same confidence .
The TTBs of route 1 with and without truncated travel time distribution are 26.89 and 26.41. And those of route 2 are 29.83 and 27.82, respectively. These imply that ignoring the effects of route travel time boundary would result in an exiguity of budget time. Travelersâ€™ riskaversion level is only dependent on their confidence level when route travel time follows a general normal distribution. In this case, travelersâ€™ riskaversion level can be calculated as 1.28. However, travelersâ€™ riskaversion level not only depends on the expected ontime arrival condition, but also is affected by the characteristics of the used routes (i.e., travel time distribution and freeflow time) when considering route travel time boundary, which is a contextbased and adaptive constant. The riskaversion level of travelers using route 1 is 1.38, and that of those using route 2 is 1.48. Then, to reveal the difference of TTR incurred by different travel time distributions, we assume travelersâ€™ TTB is a fixed constant and its quantity is 28. The TTRs of route 1 with and without truncated travel time distribution are, respectively, 0.93 and 0.95. The results indicate that neglecting the travel time boundary would overestimate the travel time reliability.
Table 1 shows the effects of scale variations of route freeflow travel time on TTB and travelersâ€™ riskaversion level. From this table, we can see that travelers adopt a longer budget time to ensure ontime arrival when they are faced with higher risk. As the freeflow travel time increases, or as longer minimum travel time is imposed on routes, the values of TTB and riskaversion level increase. Intuitively, it is clear that the growth of the minimum travel time results in a longer budget time. Thus, travelers become more aversive to risk when a longer budget time is suggested. It can be also found that travelersâ€™ riskaversion levels are different for using different routes. These results indicate that route travel time boundary has a great effect on travelersâ€™ TTBs and risk attitudes.

4. TTBBased SUE Model
This section models the effects of route travel time boundary on the performance of network equilibrium. We also present the differences between both SUE models with and without route travel time boundary. In the SUE state, no traveler can improve his or her perceived TTB by unilaterally changing route. The perceived TTBs of all used routes are the same and less than those of unused routes.
4.1. TTBSUE without Route Travel Time Boundary
In the TTBSUE model, travelers are always inclined to choose the routes with the least perceived TTB. Travelersâ€™ route choice behavior follows the random utility maximization (RUM) principle; the perceived TTB on route can be expressed aswhere is given by (14) and the random terms is assumed to be identically and independently distributed (i.i.d.) Gumbel variate, , . According to the RUM principle, the route choice probability on route at equilibrium can be written aswhere is a dispersion parameter measuring the variability in perceived TTB. A small value implies a larger perceived error compared to the shortest route. On the contrary, a large value indicates a smaller perceived error associated with the shortest route. That is, travelers would be more inclined to the shortest route when value is large enough. The equilibrium flow pattern can be obtained as followswhere is travel demand on OD pair . Apparently, the equilibrium flow pattern determined by the TTBSUE model depends upon not only the considered routeâ€™ TTB but also other alternativesâ€™ TTBs.
Mathematically, the perceived TTB under the general normal distribution can be expressed as follows.
4.2. TTBSUE with Route Travel Time Boundary
Compared to the TTBSUE without route travel time boundary, the TTBSUE with route travel time boundary uses a truncated travel time distribution to reflect the effects of route freeflow time on travelers' route choice behavior. In this case, travelers aim to minimize the perceived travel time budget with truncate travel time distribution. The new perceived travel time budget can be expressed aswhere is the perceived travel time budget under truncated travel time distribution. is given by (18) or (24). The random term is assumed to be i.i.d. Gumbel variate, , . The route choice probability on route can be written as follows.
Accordingly, we can obtain the equilibrium route flow pattern.
The perceived travel time budget under truncated travel time distribution can be expressed as follows.
4.3. The Equivalent VI Formulation
Both TTBSUE conditions can be written as follows.
So, logitbased TTBSUE problem can be formulated equivalently as the following VI problem.
Obviously, (44) can be alternatively written aswhere is the feasible flow set defined by the following.
Theorem 6. The VI problem (45) is equivalent to the equilibrium conditions of both TTBSUE models.
Proof. The proof of the theorem follows the works of Zhou et al. (2012) [30] and Bekhor et al. (2012) [31]. For the proof of the necessity, we suppose is a solution of any TTBSUE model; from the SUE condition in (37) or (41), the VI problem is satisfied naturally. Thus, any solution of both TTBSUE models is a solution of the proposed VI problem. For the proof of sufficiency, let be a solution of the VI problem. Without loss of generality, we fix a route within OD pair and construct a feasible route flow such that , , but . Then substituting them into (40), we have the following.For each effective route , , it can be deduced that . So one can obtain the following.Thus, the SUE condition in (37) or (41) is satisfied, and the solution of the proposed VI problem is also the solution of TTBSUE model.
Theorem 7. There exists at least one solution for the proposed VI problem in (45).
Proof. According to the mathematical formulation of , we know that it is a continuous function associated with , and is also a continuous function of . In addition, since is a nonempty, convex, and compact set, the proposed VI problem (45) has at least one solution.
However, the uniqueness of the solution of the equivalent VI problem cannot be guaranteed. The reason is that the solution to logitbased TTBSUE is not unique in general and the strict monotonicity of the mapping cannot be ensured.
5. Solution Algorithm
The MSA algorithm is employed to resolve the proposed equilibrium problem in this paper. And the algorithm steps are expressed as follows.
Step 1 (initialization). For given OD demand of OD pair , randomly load a feasible route flow vector . Set the maximum iteration number and the convergence tolerance . Let the iteration counter .
Step 2 (cost calculation). Determine link flow vector according to the relationship between link flows and route flows, and then calculate the mean and standard deviation of link travel time. Accordingly, we can obtain the mean and standard deviation of route travel time by (4) and (5). Thus, the travel time budget with truncated travel time distribution can be derived by (24).
Step 3 (flow assignment). Set the auxiliary route vector , which can be yielded by (40) and (41).
Step 4 (update). The route flow vector can be updated by the following.
Step 5 (checking the convergence). If the iteration counter exceeds the maximum iteration number or (51) is satisfied, the algorithm stops and the present is the equilibrium results.Otherwise, make and return to Step 2.
6. Numerical Example
6.1. A ThreeRoute Network
This section uses two transportation networks to demonstrate the proposed model and algorithm. The first network consists of three parallel routes shown in Figure 2. The characteristic values of the link parameters are given in Table 2. The parameters of BPR function in (1) are and . The total traffic demand is assumed to be fixed and it is 15,000 vehicles per hour. Without loss of generality, travelers are assumed to be homogeneous with the identical expected ontime arrival probability .

The dispersion parameter is assumed to be 1 in both TTBSUE models. Tables 3 and 4 provide the equilibrium results of both SUE models. From the tables, one can see that two SUE models give quite different equilibrium results. It implies that route travel time boundary has a great impact on travelersâ€™ route choice behavior under uncertainty. In the TTBSUE model without route travel time boundary, the riskaverse level is 1.28 no matter which route is used, which is only related to travelersâ€™ expected ontime arrival probability. In contrast, it is heterogeneous which is also affected by the used routes when considering route travel time boundary. It is plausible that travelers adopt different riskaversion attitudes for diverse routes because of the physical performance difference between different routes. From the tables, we can see that the riskaversion level is still 1.28 when using route 3. However, it is distinct when using other two routes, such as 1.64 for route 1 and 1.36 for route 2. As expected, longer travel time budget is suggested by the TTBSUE model with travel time boundary than that without travel time boundary in terms of each route. In addition, the perceived travel time budgets using different routes are identical at equilibrium, and more perceived travel time budget is caused under truncated travel time distribution.


Figure 3 depicts the effects of the expected ontime arrival probability on travelersâ€™ riskaversion levels using different routes. As increases, travelersâ€™ riskaversion levels increase for each route. In comparison, travelers using route 1 of the worst reliability adopt largest riskaversion levels to address the risk, and travelers on route 3 of the best reliability have the smallest riskaversion attitudes.
Figures 4(a) and 4(b) show the impacts of different boundary scales of route travel time on the expected travel time and risk. Specifically, the expected travel time on three routes is always increasing as the boundary scale becomes larger. Meanwhile, the most reliable route (i.e., route 3) has the longest expected travel time, whereas using the most degradable route (i.e., route 1) spends the shortest expected travel time. As the boundary scale of route travel time increases, travel time risk on three routes is also always increasing. In addition, one can see that travelers using the most reliable route are confronted with the lowest travel time risk, and those using the shortest expected travel time encounter the highest travel time risk. These imply that travelers have to make a tradeoff between expected travel time and risk. These also demonstrate that travelersâ€™ risk attitude has to be incorporated into route choice modeling. Figure 4(c) plots the variation of perceived TTB with route travel time boundary in the equilibrium state. From this figure, we can see that the perceived TTB increases constantly when boundary scale of route travel time becomes larger. All of these indicate that route travel time boundary has a great influence on travelersâ€™ travel cost (i.e., expected travel time, risk, and perceived travel time budget).
(a)
(b)
(c)
6.2. NguyenDupuis Network
In this section, the NguyenDupuis network is used to demonstrate the solvability of the proposed model with route travel time boundary under uncertainty. This network has 13 nodes, 19 links, and 4 OD pairs. The BPR performance function (, ) in (1) is adopted to compute link travel cost. Link specifications are provided in Figure 5. The potential demands are , , , and , respectively. The link capacity parameter and discrete parameter are set as and . All travelers in the test network are assumed to be homogeneous with .
First, the equilibrium results of TTBSUE with truncated travel time distribution are shown in Table 5. Without loss of generality, we only present the equilibrium results of OD (13) and OD (42). As expected, the perceived TTBs at the equilibrium on all used routes for each OD pair are equal, which is satisfied with the TTBbased SUE condition. Travelers prefer to choose the routes with shorter TTBs. It can be also observed that the travel time risk decreases with the increase in expected travel time. The riskaversion level is heterogeneous for different used routes within one OD pair, which is jointly affected by expected ontime arrival probability and route characteristics (i.e., route freeflow travel time and distribution). It is relatively larger for routes with higher risk. However, it does not mean that the larger the travel time risk, the higher the riskaversion level. As mentioned above, route travel time boundary is a major influence factor of travelersâ€™ riskaversion level.

Second, we further examine the joint impacts of expected ontime arrival probability and discrete parameter on perceived TTB in Figures 6 and 7. From these figures, we can see that the perceived TTB decreases as increases. And the greater the expected ontime arrival probability , the longer the perceived travel time budget. It implies that travelers with high requirement for arriving on time would make a longer budget time to avoid risk.
Then, the selfregulated averaging method (SRAM) proposed by Liu et al. (2009) is adopted to solve the model [32]. In the SRAM, the step size is , where
In the MSA, is enlarged by an increment of one at each iteration. It can result in the next solution keeping farther away from the optimal solution than previous one and extremely slow convergence speed. However, depends on the information between iterations to adjust the choice of step size (i.e., speeding up or slowing down) in the SRAM. Meanwhile, the choice of the step size increment parameters and is flexible, e.g., and .
Thus, we replace the step size with in (45). In addition, we assume the parameters , , , , the maximum iteration number , and the convergence tolerance . Figure 8 compares the convergence performances of SRAM and MSA. The results show that both SRMA and MSA can solve the model; the model is solvable. It is also clear that SRAM has a faster convergence speed than MSA.
7. Conclusions
This paper considers route freeflow travel time as the lower boundary of route travel time under uncertain networks. The travel time budget model with route travel time boundary is developed and derived. In such case, the budget time required by travelers is no less than that without route travel time boundary. Moreover, it has a lower travel time reliability under the same budget time. Travelersâ€™ riskaversion attitude is no longer only associated with the expected ontime arrival condition, which is selfadaptive and depends on the chosen route.
To model travelerâ€™ route choice behavior when considering route travel time boundary, logitbased stochastic user equilibrium model based on travel time budget (TTBSUE) is developed. A routebased solution algorithm is designed to solve the equivalent VI problem of the proposed model. The results indicate that route travel time boundary has a great impact on travelersâ€™ expected travel time, travel risk, and perceived travel time budget.
It should be noted that the model can be further extended to model its effects on meanexcess travel time (METT) proposed by Chen and Zhou (2010) [13] and bounded rational confidence level (BRCL) proposed by Sun et al. (2017) [33]. It can be also introduced and incorporated into dynamic travel assignment model. It would also be interesting to model travelersâ€™ riskaverse route choice decisions with multiclass users. These issues will be settled in the further research.
Data Availability
The experiment data in the submitted manuscript can be acheieved by running the program source of numerical example. The experiment program is written by Matlab 2015.
Conflicts of Interest
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Acknowledgments
Financial support provided by the National Natural Science Foundation of China (No. 51338008 and No. 51378036) is acknowledged.
References
 M. A. AbdelAty, R. Kitamura, and P. P. Jovanis, â€śExploring route choice behavior using geographic information systembased alternative routes and hypothetical travel time information input,â€ť Transportation Research Record, no. 1493, pp. 74â€“80, 1995. View at: Google Scholar
 W. B. Jackson and J. V. Jucker, â€śAn empirical study of travel time variability and travel choice behavior,â€ť Transportation Science, vol. 16, no. 4, pp. 460â€“475, 1982. View at: Publisher Site  Google Scholar
 T. C. Lam and K. A. Small, â€śThe value of time and reliability: Measurement from a value pricing experiment,â€ť Transportation Research Part E: Logistics and Transportation Review, vol. 37, no. 23, pp. 231â€“251, 2001. View at: Publisher Site  Google Scholar
 M. A. AbdelAty, R. Kitamura, and P. P. Jovanis, â€śUsing stated preference data for studying the effect of advanced traffic information on drivers' route choice,â€ť Transportation Research Part C: Emerging Technologies, vol. 5, no. 1, pp. 39â€“50, 1997. View at: Publisher Site  Google Scholar
 D. Brownstone, A. Ghosh, T. F. Golob, C. Kazuimi, and D. V. Amelsfort, â€śDrivers' willingnesstopray to reduce travel time: Evidence from the San Diego I15 congestion pricing project,â€ť Transportation Research Part A: Policy and Practice, vol. 37, no. 4, pp. 373â€“387, 2003. View at: Publisher Site  Google Scholar
 H. X. Liu, W. Recker, and A. Chen, â€śUncovering the contribution of travel time reliability to dynamic route choice using realtime loop data,â€ť Transportation Research Part A: Policy and Practice, vol. 38, no. 6, pp. 435â€“453, 2004. View at: Publisher Site  Google Scholar
 H. K. Lo and Y. K. Tung, â€śNetwork with degradable links: capacity analysis and design,â€ť Transportation Research Part B: Methodological, vol. 37, no. 4, pp. 345â€“363, 2003. View at: Publisher Site  Google Scholar
 R. B. Noland, K. A. Small, P. M. Koskenoja, and X. Chu, â€śSimulating travel reliability,â€ť Regional Science & Urban Economics, vol. 28, no. 5, pp. 535â€“564, 1998. View at: Publisher Site  Google Scholar
 G. N. Bifulco, G. E. Cantarella, F. Simonelli, and P. VelonĂ , â€śAdvanced traveller information systems under recurrent traffic conditions: network equilibrium and stability,â€ť Transportation Research Part B: Methodological, vol. 92, pp. 73â€“87, 2016. View at: Publisher Site  Google Scholar
 A. Khani and S. D. Boyles, â€śAn exact algorithm for the meanâ€“standard deviation shortest path problem,â€ť Transportation Research Part B: Methodological, vol. 81, part 1, pp. 252â€“266, 2015. View at: Publisher Site  Google Scholar
 M. Li, N. M. Rouphail, M. Mahmoudi, J. Liu, and X. Zhou, â€śMultiscenario optimization approach for assessing the impacts of advanced traffic information under realistic stochastic capacity distributions,â€ť Transportation Research Part C: Emerging Technologies, vol. 77, pp. 113â€“133, 2017. View at: Publisher Site  Google Scholar
 H. K. Lo, X. W. Luo, and B. W. Y. Siu, â€śDegradable transport network: travel time budget of travelers with heterogeneous risk aversion,â€ť Transportation Research Part B: Methodological, vol. 40, no. 9, pp. 792â€“806, 2006. View at: Publisher Site  Google Scholar
 A. Chen and Z. Zhou, â€śThe Î±reliable meanexcess traffic equilibrium model with stochastic travel times,â€ť Transportation Research Part B: Methodological, vol. 44, no. 4, pp. 493â€“513, 2010. View at: Publisher Site  Google Scholar
 A. Chen, Z. Zhou, and W. H. K. Lam, â€śModeling stochastic perception error in the meanexcess traffic equilibrium model,â€ť Transportation Research Part B: Methodological, vol. 45, no. 10, pp. 1619â€“1640, 2011. View at: Publisher Site  Google Scholar
 D. Watling, â€śUser equilibrium traffic network assignment with stochastic travel times and late arrival penalty,â€ť European Journal of Operational Research, vol. 175, no. 3, pp. 1539â€“1556, 2006. View at: Publisher Site  Google Scholar
 J. Y. T. Wang and M. Ehrgott, â€śModelling route choice behaviour in a tolled road network with a time surplus maximisation biobjective user equilibrium model,â€ť Transportation Research Part B: Methodological, vol. 57, pp. 342â€“360, 2013. View at: Publisher Site  Google Scholar
 J. Y. T. Wang, M. Ehrgott, and A. Chen, â€śA biobjective user equilibrium model of travel time reliability in a road network,â€ť Transportation Research Part B: Methodological, vol. 66, pp. 4â€“15, 2014. View at: Publisher Site  Google Scholar
 C. Sun, L. Cheng, S. Zhu, F. Han, and Z. Chu, â€śMulticriteria user equilibrium model considering travel time, travel time reliability and distance,â€ť Transportation Research Part D: Transport and Environment, 2017. View at: Google Scholar
 H. Shao, W. H. K. Lam, Q. Meng, and M. L. Tam, â€śDemanddriven traffic assignment problem based on travel time reliability,â€ť Transportation Research Record, no. 1985, pp. 220â€“230, 2006. View at: Google Scholar
 H. Shao, W. H. K. Lam, and M. L. Tam, â€śA reliabilitybased stochastic traffic assignment model for network with multiple user classes under uncertainty in demand,â€ť Networks and Spatial Economics, vol. 6, no. 34, pp. 173â€“204, 2006. View at: Publisher Site  Google Scholar
 W. H. K. Lam, H. Shao, and A. Sumalee, â€śModeling impacts of adverse weather conditions on a road network with uncertainties in demand and supply,â€ť Transportation Research Part B: Methodological, vol. 42, no. 10, pp. 890â€“910, 2008. View at: Publisher Site  Google Scholar
 Z. Zhou and A. Chen, â€śComparative analysis of three user equilibrium models under stochastic demand,â€ť Journal of Advanced Transportation, vol. 42, no. 3, pp. 239â€“263, 2008. View at: Publisher Site  Google Scholar
 B. W. Y. Siu and H. K. Lo, â€śDoubly uncertain transportation network: degradable capacity and stochastic demand,â€ť European Journal of Operational Research, vol. 191, no. 1, pp. 166â€“181, 2008. View at: Publisher Site  Google Scholar
 H. Sun and Z. Gao, â€śStochastic Traffic Equilibrium Based on Travel Time Robust Reliability,â€ť Journal of Transportation Systems Engineering and Information Technology, vol. 12, no. 2, pp. 76â€“84, 2012. View at: Publisher Site  Google Scholar
 S. V. Ukkusuri, T. V. Mathew, and S. T. Waller, â€śRobust transportation network design under demand uncertainty,â€ť ComputerAided Civil and Infrastructure Engineering, vol. 22, no. 1, pp. 6â€“18, 2007. View at: Publisher Site  Google Scholar
 Y. Yin, S. M. Madanat, and X. Lu, â€śRobust improvement schemes for road networks under demand uncertainty,â€ť European Journal of Operational Research, vol. 198, no. 2, pp. 470â€“479, 2009. View at: Publisher Site  Google Scholar  MathSciNet
 C.Y. Yan, R. Jiang, Z.Y. Gao, and H. Shao, â€śEffect of speed limits in degradable transport networks,â€ť Transportation Research Part C: Emerging Technologies, vol. 56, pp. 94â€“119, 2015. View at: Publisher Site  Google Scholar
 X. Xu, A. Chen, H. K. Lo, and C. Yang, â€śModeling the impacts of speed limits on uncertain road networks,â€ť Transportmetrica A: Transport Science, vol. 14, no. 12, pp. 66â€“88, 2018. View at: Publisher Site  Google Scholar
 X. Xu, A. Chen, L. Cheng, and H. K. Lo, â€śModeling distribution tail in network performance assessment: a meanexcess total travel time risk measure and analytical estimation method,â€ť Transportation Research Part B: Methodological, vol. 66, pp. 32â€“49, 2014. View at: Publisher Site  Google Scholar
 Z. Zhou, A. Chen, and S. Bekhor, â€śClogit stochastic user equilibrium model: formulations and solution algorithm,â€ť Transportmetrica, vol. 8, no. 1, pp. 17â€“41, 2012. View at: Publisher Site  Google Scholar
 S. Bekhor, C. Chorus, and T. Toledo, â€śStochastic user equilibrium for route choice model based on random regret minimization,â€ť Transportation Research Record, no. 2284, pp. 100â€“108, 2012. View at: Google Scholar
 H. X. Liu, X. He, and B. He, â€śMethod of successive weighted averages (MSWA) and selfregulated averaging schemes for solving stochastic user equilibrium problem,â€ť Networks and Spatial Economics, vol. 9, no. 4, pp. 485â€“503, 2009. View at: Publisher Site  Google Scholar  MathSciNet
 C. Sun, L. Cheng, and J. Ma, â€śTravel time reliability with boundedly rational travelers,â€ť Transportmetrica A: Transport Science, vol. 14, no. 3, pp. 210â€“229, 2018. View at: Publisher Site  Google Scholar
Copyright
Copyright © 2018 Lei Zhao 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.