You are on page 1of 24



1. ABSTRACT This paper presents an empirical comparison of alternative methods for computing user equilibrium on large regional transportation networks. Specifically, it examines the solution characteristics, convergence behavior and associated computing times of link-based, origin-based, and path-based methods. The latter two methods use more memory and are computationally more demanding, but have been touted in the literature as converging more rapidly than the link-based Frank-Wolfe method or variants thereof, that are commonly employed in planning software. The principal motivation in searching for improved methods is achieving more rapid and/or tighter convergence in the computation of equilibrium. Greater convergence is needed for accurate forecasting the impacts associated with road and public transport projects and affects nearly all aspects and components of transportation models as well as being a major determinant of their internal consistency. Congested travel speeds are typically used to compute trip distribution and mode choice, and these speeds will be incorrect if a satisfactory traffic assignment is not achieved. Due to long computational times, many models are insufficiently calibrated and converged for forecasting purposes. This problem is partly the result of and is compounded by the slow convergence of the Frank-Wolfe algorithm. In order to perform this research, origin-based and path-based algorithms were coded following the existing literature and tested on networks on large networks of the type that are representative of regional travel demand forecasting efforts for major metropolitan regions. The algorithms coded were those described by Chen et al.(2002) and Dial (2006). An executable version of origin-based traffic assignment code written by one of its proponents (Bar-Gera, 2002) and made freely available for research was also tested. The new algorithms were compared with the current production version of the traffic assignment in TransCAD which is based upon the Frank-Wolfe algorithm and has been shown to be faster than other commercial implementations in the United States. Concurrent with this research, the F-W traffic assignment in TransCAD was multi-threaded leading to speedups in computation proportionate to the number of central processing units and/or CPU cores available. This raises the bar for new assignment algorithms, especially if they are less suitable than F-W for multi-threading.

©Association for European Transport and contributors 2006

The initial tests revealed that the Bar Gera executable and the Chen et al. pathbased method described in the literature would need improvements before they could be competitive in performance computing user equilibrium with existing link-based codes. The path-based method did not converge on medium to large size networks. We made modifications to the gradient search, so that it converged albeit not usually as efficiently as other methods. The Bar Gera originbased method converged tightly, but only after very long computational times rendering it impractical for use by planners. However, our implementation of an origin user equilibrium method developed in conjunction with Robert Dial and based upon his algorithm B demonstrated superior performance in reaching high levels of convergence in significantly less computing time than F-W. It also can reach a tight equilibrium even more quickly from a prior solution resulting in much lower computing times for models with feedback and most forecasting tasks. Our testing on a realistic planning model trip table and network suggest that the origin user equilibrium can be deployed by practitioners with immediate benefits in terms of reduced computing times and more tightly converged models. 2. INTRODUCTION This paper reports on an ongoing empirical investigation of alternative traffic assignment methods. Over the last decade, new methods for computing user equilibrium traffic assignments have been researched and have resulted in published claims of superiority over the most widely used methods. Yet these new methods have seen little if any use by practitioners who develop and apply travel demand forecasting models. In this paper, we present an assessment of path-based and origin-based assignment methods and compare their computational performance with the conventional link-based, Frank-Wolfe method in widespread use. The principal motivation in searching for improved methods is achieving more rapid and/or tighter convergence in the computation of equilibrium. Greater convergence is needed for accurate forecasting the impacts associated with road and transit projects and affects nearly all aspects and components of transportation models as well as being a major determinant of their internal consistency. Congested travel speeds are typically used to compute trip distribution and mode choice and these speeds will be incorrect if a satisfactory traffic assignment is not achieved. Due to long computational times, many regional models are insufficiently calibrated and converged for forecasting purposes. This problem is partly the result of and is compounded by the slow convergence of the Frank- Wolfe (FW) algorithm that is used to compute user equilibrium in the most commonly used software packages. In the course of examining the traffic assignment components of many regional models in the U.S. and elsewhere, we identified low levels of convergence, improper measures of convergence, and methods that either do not attempt to calculate user equilibrium or fail to do so correctly. In this paper, we also hope to

©Association for European Transport and contributors 2006

Leurent. we focus exclusively on the computation of a static user equilibrium (UE) as defined by Wardrop’s condition that all used paths for each origindestination pair have the same minimum cost. bicriterion UE (Dial. Accordingly. the origin user equilibrium approach that implements a version of Dial’s algorithm B can provide superior convergence performance and can be deployed in the near term to achieve reduced computing times. a series of all or nothing assignments are performed and flows are combined using weights derived from a line search that attempts to minimize the linearized objective function. BACKGROUND ON USER EQUILIBRIUM In this paper. no traveler can switch to a shorter path and improve his or her travel time. Leblanc et al. These computers provide significant speedups for multi-threaded implementations of traffic assignment algorithms. Results from Boyce et al. All of the link flows emanating from all ©Association for European Transport and contributors 2006 . and thus have a bearing upon the comparison of alternative methods. or dynamic models may be behaviorally more realistic. The literature to date has not addressed the properties of these solutions empirically and there are open questions about how similar the solutions are to those currently obtained. Our focus is on achieving computational rather than representational accuracy. or both. In other words. We also attempt to provide some information about the nature of the solutions that come from alternative methods. In congested networks. 1996. greater convergence. under assumptions of route costs that are additive link costs and link costs being simply a (continuously differentiable. it should be a computationally stable method of generating forecasts. Beckmann et al. inexpensive computers with either two CPUs or two cores or both are now widely available. we report results from two generations of hardware. nondecreasing) function of link flows. Bernstein (1990) has shown that UE has good stability with respect to small perturbations. (1975) proposed using the Frank-Wolfe (FW) method for computing equilibrium that was implemented in UTPS and which underlies most planning software implementations in use today. computing environments continued to evolve and improve. In particular. that the traffic assignment problem could be formulated as a minimization problem. (2004) provide empirical support for this conclusion. 3. 1996). if a tight equilibrium solution can be generated. Other traffic assignment models such as stochastic user equilibrium (Sheffi. 1982). consequently. This method has been extended to handle turn penalties and multi-class assignment and will be available commercially in the near future. In the FW method. Achieving much tighter convergence in an equal or lesser amount of computing time would be a breakthrough for practitioners. Based upon our findings. During the course of this investigation. user equilibrium is characterized by the use of many paths for many OD pairs. (1956) demonstrated.provide some better guidance on convergence for practicing modelers.

the difference between the total cost of the current User Equilibrium. solution ( cUE ) and the total cost of the All-or-Nothing. a stopping criterion is required. As a are updated during each iteration. Note that if the minimum path travel time between each OD pair does not change. Maximum flow change  xin − xin −1   max   xin   3. Rose et al. (1985) lists several reasonable stopping criteria that might be used and that are shown below: 1. The process is repeated for a specified number of iterations or until some stopping criterion is met. As a result. Change or percent change of the objective function  z n − z n −1      zn   2. the Wardrop condition is satisfied since there are no lower cost alternatives for any traveler. the order in which the origins are processed is not of consequence. UE. at equilibrium. at iteration n Volume delay function Demand ©Association for European Transport and contributors 2006 . therefore. AON. but the cost at the AON solution (which is always available since it determines the direction of search for the next iterate) serves as a lower bound on the equilibrium solution for the current iteration. Average Excess Cost ∑ xUE ⋅ c( xUE ) − ∑ x AON ⋅ c( xUE ) ∑ OD where: n i z xin c(⋅) OD Iteration number Link index Objective function Flow at link i . solution ( c AON ) is zero and the difference is. a natural measure of convergence. Relative gap ∑ xUE ⋅ c( xUE ) − ∑ x AON ⋅ c( xUE ) ∑x UE ⋅ c( xUE ) 4. Obtaining the value of the objective function requires an extra calculation. Since the solution algorithm to the Traffic Assignment Problem is iterative in nature.

Second. too. 1994). Packages that rely on incremental methods or limit the number of iterations will typically fail to reach good solutions. there is little empirical evidence of efficiency gains on the large networks that are used in planning applications. the “GAP” reported is completely different from the relative gap measure defined above and is computed from successive UE iterates without consideration of the AON solution. most models initiate the first round of trip distribution and mode split model application using congested highway times. This is the measure that we rely on for comparisons among alternative methods. Apart from these measures. but some may be potentially misleading. The large amount of time required to compute equilibrium on large metropolitan networks is one reason that regional traffic assignments do not achieve good convergence. Bar Gera uses “average excess cost” which is the difference between each route’s time and the time of the shortest route weighted by the route’s flow (Boyce and Florian. While there is a large literature on how to speed up FW and assignments in general using other link based methods such as simplicial decomposition (Patrikson. One exception is the PARTRAN method which reportedly has some advantages and is implemented in the EMME/2 package (Florian et al. other stopping criteria are often encountered and can be useful. which is used to compute some more elaborate traffic assignment models. and this. 1987). This measure does not indicate how far from equilibrium a solution may be. These measures are not comparable to those employed here and greatly overstate the degree of convergence obtained. is probably an effective if not a more effective measure. while efficient early on. However. In his papers on origin-based assignment. 1985). Another problem is that some packages do not provide convergent algorithms for user equilibrium traffic assignment although they purport to do so. congested equilibrium travel times are a key calibration input for parameter estimation for distribution and mode choice models. The method of successive averages (MSA). This is a fairly sensitive measure of convergence and is superior to many other stopping criteria such as simple functions of the differences between assignment iterates (Rose et al.. is generally considered to be much slower than FW for congested networks. then trip distribution and mode choice will be affected.. 2005). The maximum link flow change between iterations can help understand the general degree of convergence. This has the deepest ramifications for almost all aspects of travel forecasting models. A practical problem for modelers is that the FW algorithm.The “relative gap” is the aforementioned difference between the cost of the current UE solution and the cost of the AON solution divided by the cost of the current UE solution. First. Good convergence is also ©Association for European Transport and contributors 2006 . In some planning software. exhibits slower convergence as it progresses toward equilibrium. Dial(2006) suggests using a tolerance value that is based upon the difference between the longest and shortest used paths. We also use this measure in comparing the performance of alternative assignment methods. it is not computable for FW solutions so we do not make use of it in our comparisons. If convergence is poor.

but potentially renders the results order dependent.g. and unless terminated because of the stopping criterion. Lee.1. Chen et al. In any case. The machine used was a 256MB RAM Sun workstation. The path-based algorithm tested is the method introduced by Jayakrishnan et al. (2002). which is quite efficient in its memory requirements... ALTERNATIVE ASSIGNMENT METHODS Our motivation in examining alternative methods is to find more rapidly converging algorithms. and Jayakrishnan (2002). Bar-Gera. reported results was the Advance sketch planning network for Chicago. We consider two classes of methods--path based and origin-based methods. the algorithm performs column generation (i. both of which require more memory than the FW algorithm. and Boyce et al. Bar-Gera and Boyce (2002). This is generally thought to be faster because flow and cost updates are relatively inexpensive compared to column generation. there is a literature that has touted these methods as having superior performance (e. 4. 4. This is more effective than FW which has trouble moving flow completely away from inefficient paths. (2002). Dial (1999). and Dial (2006). (2004). further evolved by Chen and Jayakrishnan (1998). After an initialization with an all-or nothing assignment. and described in Chen.e.. it equilibrates the flow over the previously identified paths. The Path-based Method Although path flow solutions are not unique. Path based methods require keeping track of all utilized routes which. Consequently. The origin-destination pairs are processed one at a time. it tests for another shortest path for each O-D pair). Of course it is just as feasible to save all the paths generated by any other method including F-W and to use compression to store paths more efficiently. and a relative gap of . Moreover. This method has the advantage that paths can be dropped if they have zero flow. The largest network for which Chen et al.001 was used as the measure for a ©Association for European Transport and contributors 2006 . some proponents erroneously believe that path-based methods are attractive because they lead to a direct characterization of the routes that are utilized in a traffic assignment solution rather than just the total link flows.needed to make feedback worthwhile and effective. This equilibration is performed using a gradient search to move flow from longer used paths to shorter paths. The rapid advances in computing power available to demand forecasters and the enormous increases in available memory and storage space make consideration of these methods feasible now when they were not before. many models and forecasting applications fail to achieve good convergence and give misleading results. for a large problem could be on the order of 100 million paths or more. our focus here is simply on the computational efficiency of the most effective path based algorithms in computing user equilibrium link flows. (1994).

1999. However. 1987. In this research. The idea behind origin-based assignment is that the equilibrium solution for each origin is an acyclic graph (Jansen and Zozaya-Gorostiza. Dial (2006) refers to his algorithm B as a “path-based algorithm that obviates path storage and enumeration. 2006). Origin-based methods maintain acyclic solutions by processing of origin “bushes” or subnetworks (Dial. Even if this is overkill. 2006). It is also related to the use of poor measures of convergence that give a false sense of security and a lack of appreciation of convergence issues on the part of practicing planners. We didn’t have access to the details of Bar-Gera’s implementation and therefore could not assess whether or not it could be sped up.” but in our classification we consider it an origin equilibrium method. Instead of keeping all the paths in memory. Origin approaches use this subnetwork. 4. it was much slower than our FW implementation for less convergent target gaps. Bar-Gera’s algorithm has been able to compute solutions to tiny gaps. but much less memory than path-based methods. convergence. When we first noted this. it makes it possible to compare planning solutions with more highly converged ones. 1999. we tested larger networks and tighter Origin-based Methods Origin-based assignments have been proposed by Bar-Gera (1999) and Dial (1999. have more efficient shortest path calculations than FW. we wondered whether or not this achievement was of any practical utility. Consequently. Origin-based methods for UE require more memory than FW. Boyce and Bar-Gera deserve credit for calling attention to the poor convergence associated with most planning models. Since machines differ greatly in terms of performance. and prohibit flow from links that are part of cycles giving greater computational efficiency. this taking more than 2 days to compute. Dial. 1999). we focused on implementing an origin-oriented traffic assignment based upon Dial’s algorithm B. we found the published running times rather high relative to our experience with UE in existing models.converged solution. one keeps the solution for each origin. In any case. He reports average excess costs on the order of 1e -11. ©Association for European Transport and contributors 2006 . we did some joint testing with him and concluded that while his algorithm might reach low gaps.2. The temptation to control execution times has lead to insufficient convergence for the intended purposes of planning models. This multiplies the memory use by a factor relating to the number of origins although compression can be utilized. this is not necessarily an algorithmic problem. This addresses a major weakness of FW which has trouble removing cycles once they arise. In reviewing Bar-Gera’s work. BarGera.

We refer to our implementation as origin user equilibrium or OUE to distinguish it from Bar-Gera method. The initial versions of OUE and our F-W codes were single-threaded. This led to much improved convergence. Instead.5. This underscores the need for improved convergence as well as shortened computational effort. However. Examination of the path-based method revealed that the gradient projection search algorithm was not very efficient and could be problematic. We took the conservative approach and the OUE is less effectively multithreaded for now. although not to the same degree. Dial (2006) is the definitive reference for algorithm B. In our experience. Multi-threading is a great way to speed up computations when PCs with multiple CPUs or CPUs with multiple cores are available. THE TESTING APPROACH To test these methods. Given the trend toward multi-class. algorithms that take a long time to compute are not attractive to practitioners. ©Association for European Transport and contributors 2006 . we focused on their ability to give good results and converge within a modest amount of computing time. but it was clear that memory requirements and computing times might be too big for the largest networks in use. we made some modifications that improved the step size calculation. After achieving good results with our algorithm B-based OUE. and the method was not straightforward to code from the descriptions in papers. it is not uncommon for the effective number of assignments performed to be multiplied by the number of classes and the number of feedback loops. IMPLEMENTATION We implemented our own version of the Chen et al. The modification was necessitated to achieve convergent behavior on the large networks that we were using. Then they were multi-threaded. single class assignment. for algorithms which are order-dependent such as OUE the same cannot be said. In the case of the F-W algorithm. path-based traffic assignment in the C programming language. This is important if the same results are to be obtained when different computers are used. Our path-based algorithm is the same as that referenced above. This often leads to computing times on the order of 25 times or more than that for a single. Consequently. both of which are requirements for a production code. The Bar Gera source code was not available to us. but with a modified gradient search procedure. However. multi-mode assignments and use of feedback loops. 6. we worked with Robert Dial to implement his algorithm B and created an alternative origin user equilibrium method following his pseudo-code in most respects. multi-threading is relatively straightforward. extra care is necessary if one wishes to ensure that the results are numerically the same for multi-threaded versus single-threaded implementations. we then created a version that can handle multiple user classes and turn penalties.

It is harder to control for problem characteristics such as congestion levels and volume delay functions. we have minimized the confounding effects of memory availability and the speed of hard disk or RAM access can affect reported running times significantly especially for small networks where reads and writes might account for a disproportionate amount of total computing time.2 GHZ dual core Athlon with 2 gigabytes of RAM also running XP Professional. We also considered memory utilization which can determine the feasibility of applying some of these methods to the largest planning networks currently in use.We were particularly interested in how much convergence could be achieved in half an hour or less which is about the limit (per class) of interest to planners who do multi-class assignments. Early in the development of traffic assignment methods. we used BPR functions but we experimented with BPR parameters that varied across networks and across the different types of links within the same network. the relationship between travel volumes and network capacity influences the rate of convergence of all algorithms.0 GHz Pentium M PC with 2 gigabytes of RAM running under Windows XP Professional. and one of the main advantages of Frank-Wolfe method is that it has minimal memory requirements. it is twice as fast on a dual core machine and four times as fast on a two CPU machine with two cores per CPU.S in less than one-half hour using a released version of TransCAD on a fast PC in 2005.0001. However. the slower is the convergence that will be obtained from the FW algorithm and presumably from other methods. the value recommended by Boyce et al. and which we initially thought might be the lowest value of practical rather than theoretical interest. Consequently. (2004). The traffic assignment used for comparison is the released TransCAD version which is multi-threaded. we used a 2. Generally speaking. Problems that don’t fit in memory can be solved at the expense of execution time associated with reading and writing to disk. We developed a rigorous comparison methodology for performing our tests. Volume delay function properties can also affect convergence. This involved standardizing on computing environments and test networks and adopting consistent measures of effectiveness. This is a fast single core inexpensive computer. Subsequently we used a 2. although now during the Summer of 2006 much faster computers are available. By standardizing the computing environment. In all of our comparisons reported here. This machine is quite representative of current offerings. minimum memory utilization was a major consideration in designing algorithms. For our initial testing. we cannot be sure that some of our ©Association for European Transport and contributors 2006 .001 or better which is achievable on virtually all large networks in the U. We were then interested in the ability to achieve a relative gap of . Memory requirements are also important because there are definite tradeoffs in execution speed and use of memory. The greater the flow relative to capacity. Our feeling was that the convergence range of interest was a minimum relative gap of .

Bad test cases can lead to bad conclusions. We used a variety of test networks in our research. To avoid undue complications. The Chicago network has 39. and average excess costs were computed for each traffic assignment solution. This measure has a direct interpretation and is potentially more stable because the denominator does not change as a function of convergence or the number of iterations. The Washington D.conclusions would be different if steeper or flatter volume-delay functions were utilized. and 1489 zones.286 nodes.000 nodes. single precision floating point calculations. we extended our comparisons to a variety of other networks in use in the U.421 links. Other permutations are possible.. we believe. This is the amount that the link flow solution changed on the last iteration and is an indication of how much of change might be expected if another ©Association for European Transport and contributors 2006 . the roadway trip table when assigned matches ground counts closely. Additional precision might be warranted for super high convergence levels. Bar-Gera uses average excess cost which has the same numerator as the relative gap but a different. Concerned that the results still might be problem dependent. D. for major metropolitan areas. fixed denominator which is the number of trips.000+ links. Bigger networks have many more alternative paths and much greater computational complexity. 7. While not reported here. The Philadelphia network has just over 40.000 links. Numerical treatments range from using integers multiplied by 100. we also use Bar-Gera’s Philadelphia network and a network from a model that we recently calibrated for the region surrounding and including Washington. This fact has been known for a very long time (Rose et al. the objective function. Networks attributes such as link costs may also differ in their numerical representation even when they supposedly start from the same input values. We also compute the maximum flow change link flow change at different levels of convergence.C. and 2523 zones. or doubles for arithmetic calculations. Moreover. The excessive use of the Sioux Falls and other small networks. has generally been misleading with respect to algorithmic performance. The latter network was developed by us for a demand modeling project and is highly accurate in terms of geography. and 1771 zones. 1985). we used double precision floating point for computing link flows. but we focused most of our initial tests on the large Chicago regional planning network that Bar-Gera used and has made available to researchers. regional network is much larger and has 62.S. almost 13. more than 13. In the tests reported in this paper.C. the same general pattern of findings was obtained.000 nodes. MEASURES OF EFFECTIVENESS The relative gap. 22.

001 and . Tables 2 and 3 show the maximum flow change and the average excess costs respectively for each method and level of convergence. To reach a relative gap of . This is more than three times faster than UE and seven times faster than OBA. the origin user equilibrium assignment based on Dial’s algorithm B needs less than 27 minutes. There can be a further ambiguity in the use of a specific relative gap. The excellence convergence of OUE is also illustrated in Figure 2 which plots the objective function (i. We use the latter measure. On this network.0001 than FW. Note that for this network and flow matrix. the forecast may not be at all accurate. 8. the Beckmann value) as a function of CPU time. which illustrates the rates of convergence for the standard. Intel Pentium M. but the algorithm B-based OUE lived up to its promise and converges rapidly and descends more quickly to lower gaps. note that the AEC measure favors the origin assignment methods at all levels of AEC. OUE is much faster. It needs less than 10 minutes whereas UE takes more than 15 minutes.001 and nearly three hours to reach . One can compute the time it takes to reach this value with rounding or the first iteration that is strictly below the target value.01. Bar-Gera’s OBA is slower still. If the maximum flow change is large relative to the potential impact of a project. To get to . The Bar-Gera executable is especially slow taking 66 minutes to reach . As shown in Table 1.0001 might provide good results for estimating impacts of projects. The path-based method has steeper convergence than UE but starts slower and is much slower to reach a relative gap of . It is four times faster than Bar-Gera’s OBA and about two-thirds of the time taken by OUE. To get a feel for each level of convergence. RESULTS The results of our first tests on the Chicago network are shown in Figure 1. The path-based method is consistently faster than Bar-Gera’s OBA code.iteration were run. The results suggest that relative gaps between . ©Association for European Transport and contributors 2006 .0001 or better. Also. the equivalent values for average excess cost as the relative gap are a tougher convergence criterion than the relative gap.e. which we feel is an excellent result.0001.001. the path-based method is almost competitive with UE but only to a relative gap of . UE is the fastest method to obtain a relative gap of .0001. Frank-Wolfe-based TransCAD UE and the three alternative assignment programs run on the single core 2GHz.

AMD dual Athlon. due to multi-threading the dual core solution is obtained virtually twice as quickly as that for the single core. Further investigation will be required to understand this result. OUE takes less than one-third the time as Frank-Wolfe and the 21+ minutes taken is quite practical for use. As is evident.2 GHZ. a similar but not identical convergence pattern was obtained for the Philadelphia network and trip table.001 or lower. the most representative of networks in use today.0001. OUE is twice as fast to .0001 as FW and the path-based algorithm is not competitive. On the two-core. so our conclusion is that OUE is best reserved for tight convergence.2. As shown in Table 10. this is a faster machine than the one used previously. While OUE performs excellently. Because it has two cores. UE (FW) is the faster algorithm.001. Dual Athlon.001 than for the other methods suggesting that algorithm B is doing more work at this stage. As shown in Table 4. An interesting pattern emerges from Table 5. on this network.1.0001. although there is still some gain from the dual core configuration for these methods. Differences between FW UE and OUE Link Flows ©Association for European Transport and contributors 2006 . This is reflected in the lesser amount of CPU time required to reach convergence for each method. the following results were obtained for Chicago.C. Recall that while both the UE and the OUE are multi-threaded. It has different BPR functions for different types of roadways following Highway Capacity Manual guidelines and further calibration to traffic counts. The final network tested was. to our way of thinking. to a relative gap of . As illustrated in Table 7. The small max flow change for UE at .8. area for the Maryland-National Capital Park and Planning Commission. Our suspicion is that this result is due to lower congestion in the Philadelphia network. Figure 3 shows a comparison of the performance of the standard TransCAD UE on this computer using a single core and two cores. On a four core machine. Multithreaded Results The remainder of the results that will be presented come from runs on the twocore. To reach . but OUE becomes better at relative gaps of . the UE is multithreaded more effectively. However. 8. it appears to be an easier problem to solve.01. FW would be even faster.0001 reflects the slow convergence of Frank-Wolfe. the path-based method is the fastest to a relative gap of . the multi-threaded FW is faster to a relative gap of . 2. Although the traffic assignment problem for Philadelphia is of similar size to that for Chicago. Neither the Bar Gera code nor the path-based codes were explicitly multi-threaded. It comes from a well-calibrated travel demand model that Caliper developed for the Washington D. Note that the max flow change at each reported relative gap is progressively larger for OUE above a gap of .

00001 gap is reasonably close to the OUE solution at a . The maximum difference in link flows is 172 trips. He presented results by factoring the Chicago trip matrix up and down by various percentages.C. In these situations. 8.Practitioners are notoriously slow to change their modeling methods. regional network with an OUE solution that was converged to a relative gap of . The second point is that the link flow solution to a UE traffic assignment converged to a relative gap of .01 or 1 percent is quite far away from a highly converged solution.8 percent. the differences in link flows can be quite substantial. Dial referred to this as a pivot-point assignment.001 which probably exceeds the tolerance of nearly all large regional models in the U. We would not expect that the link flows would be identical at any early point along the way to a very tight equilibrium. This is that having once obtained a good solution and saved the results.0000001. it should be much faster to compute a new equilibrium for a similar problem. Warm Start Solution Performance An important benefit cited by Dial (2006) was a key aspect of our motivation for exploring alternative methods.3. and they will undoubtedly wonder how different the solutions are that generated from the OUE method from those that they currently compute. the maximum link flow difference is nearly 300 trips. If this is the case. As far as we know. Normally. The UE FW result at a . D.0000001 relative gap with an RSME of just 2. Even at a gap of . this is a much tighter solution than has ever been used in a deployed forecasting model. Because of potential order dependence. To address this question. Table 13 shows the percent root mean square error (RSME) between the UE and OUE link flows. The reason is that all traffic assignment solutions bear the imprint of the algorithm that is employed to compute them. origin user equilibrium methods are not recommended for traffic assignments unless tight gap tolerances are achieved. The first point to note is that highly converged UE and OUE link flows are similar. This is what we would expect if both algorithms converge to the same unique equilibrium point as they should. OUE would reduce the time required when feedback loops are run. The latter computation took slightly more than 5 hours.. In all cases the savings were on the order of 50 percent or better to get to a relative ©Association for European Transport and contributors 2006 . the network structure does not change even though the trip table does. For this traffic assignment. we compared different FW UE solutions for the Washington. when using a F-W based algorithm each traffic assignment takes roughly the same amount of time as the first.S.

e.. The computational savings illustrated here are dramatic and suggest that OUE will confer significant benefits to practitioners.S. 9.0001 and saved the results. CONCLUSIONS From the empirical comparisons presented. This took 41 minutes on the dual core Athlon. In theory. Further study is warranted to establish the properties of the solutions generated by OUE and path-based assignment methods and to provide empirical guidance for how small the relative gaps should be. using a warm start reduced computing times for traffic assignments to one-fifth of those associated with a regular assignment with a cold start. The warm start properties of OUE are particularly attractive for computing models with feedback and for forecasting project impacts. As shown in Table 14. We sought to confirm these results on a different network and offer a stiffer test to examine the warm start behavior of OUE. reflecting a road improvement.0001 or lower for virtually all large models in the U. We also doubled the capacities on two suburban links. we computed a traffic assignment on the regional DC network to a relative gap of . Multithreading is of immediate benefit in accelerating Frank-Wolfe algorithms and will undoubtedly become vital for other assignment methods as well. In travel forecasting more generally. There seems to be little risk in deploying OUE because the results will be similar to those obtained by current methods if low relative gaps are achieved. we conclude that the origin user equilibrium method based upon Dial’s algorithm B offers the prospect of much tighter convergence than Frank-Wolfe-derived UE traffic assignments and greatly reduced computing times for very small relative gaps. more congested) traffic assignment problem. changes are made in the network and there are also changes in the trip table. with a warm start from the saved results. Fortunately. We then randomly increased the number of trips in the origin-destination matrix by 5 percent. To examine these issues. Also. ©Association for European Transport and contributors 2006 . it is interesting to note that the objective function is lower for the result that begins with the warm start suggesting that it is a better solution than the one from the cold start. OUE makes it feasible to calculate traffic assignments with gaps of . OUE may also be able to reduce the time taken to generate the first traffic assignment for a forecast (when the base case solution has been run and saved) as well as subsequent ones in feedback loops. In the empirical example we of .0001 as this was a harder (i.0001. OUE was able to compute a new traffic assignment to the same relative gap in 12 minutes. A cold start on the same traffic assignment problem with increased demands took 49 minutes to reach .

MA Chen. H and Boyce. C. D. We thank Anthony Chen and Der Hong Lee as well as Shlomo Bekhor for assistance in accessing their research on path-based methods. (1999) Entropy Maximization in Origin-based assignment.. 2004 Boyce. Ralevic-Dekic.D. D. D. We would also especially like to thank David Boyce and Hillel Bar-Gera for many discussions about originbased assignments and for providing access to their research and Hillel BarGera’s executable code for OBA.509-518 ©Association for European Transport and contributors 2006 . University of Pennsylvania. Transportation and Traffic Theory 397-415 Beckman. New Haven Bernstein./Feb. ASCE. 2005 Caliper Corporation (2005) Travel Demand Modeling with TransCAD. Version 4. Transportation Science 36 (4). H. (1990) Programmability of continuous and discrete network equilibria. unpublished draft paper. A. D. C.tighter convergence of user equilibrium models will be more easily achievable in the future. (2002) Computational study of stateof-the-art path-based traffic assignment algorithms. Jan.8. and Florian. (2005) Workshop on traffic assignment with equilibrium methods. Transportation Research Board Meeting. Yale University Press. Journal of Transportation Engineering.. January 20.. and Bar-Gera. ACKNOWLEDGMENTS The authors are indebted to Robert Dial for his work with us on path and originbased approaches and for mentoring us on traffic assignment algorithms. January 9. D-H. Mathematics and Computers in Simulation 59. and Jayakrishnan. H (2004) Convergence of Traffic Assignments: How much is enough?. McGuire. 2005 Boyce. 398-417 Bar-Gera.. R. Boyce. D. Newton. 11. REFERENCES Bar-Gera. Ph. (2002) Origin-based algorithm for the traffic assignment problem... 10. M. D. (2005) Understanding the solution of the sequential procedure with feedback. M. thesis. and Winsten. Lee. (1956) Studies in the Economics of Transportation.

. An efficient implementation of the “PARTRAN” variant of the linear approximation method for the network equilibrium problem. 75-83 LeBlanc. Elsevier Science. Transportation Research B. H. 261-274 ©Association for European Transport and contributors 2006 .. 319-339 Leurent. (1996) The theory and practice of a dual criteria assignment model with a continuously distributed value-of-time. (1996) Bicriterion Traffic Assignment: Basic theory and elementary algorithms. Volpe National Transportation Systems Center. (1994) The Traffic Assignment Problem.. L.455-477. Utrecht Rose. F. E. pp. W. forthcoming Florian.. MA July 25. 22B No. 309-318 Patrikson. (2006) A path-based user-equilibrium traffic assignment algorithm that obviates path storage and enumeration. S. R. 4. (1987) The problem of cyclic flows in traffic assignment. Daskin. Prashker. Tsai. (1988) An examination of convergence error in equilibrium traffic assignment models. Cambridge. in J. J. F.. and Pierskalla. (1975) An Efficient Approach to Solving the Road Network Equilibrium Traffic Assignment Problem.. and Spiess. 21B. C. Networks 17.299-310 Jayakrishnan. W. and Rajadhyaksha. R. Lesort (ed) Transportation and Traffic Theory. R. 1999 Dial.. M. (1987). Guelat. G. Transportation Research. and Zozaya-Gorostiza.K... M. VSP. Transportation Research B. B.. Morlok.Dial. Ltd. Transportation Research Vol 9. Vol. Transportation Research Record 1443.93-111 Dial. (1999) Algorithm B: Accurate Traffic Equilibrium (and How to Bobtail Frank-Wolfe. J. Oxford Jansen. and Koppelman. R. Transportation Science 30/2. (1994) A faster path-based algorithm for traffic assignment. M..B..

E-07 Time (min) ©Association for European Transport and contributors 2006 .E-03 1.E-04 O_UE OBA FW Path Based 1. 2GHz Pentium M Relative Gap 0 1.Chicago Network.E-06 1.E-05 Gap 1.FIGURE 1: CONVERGENCE versus CPU TIME .E-02 50 100 150 200 1.

CPU Time. 2GHz Pentium M 26000000 25980000 25960000 Objective Function 25940000 25920000 25900000 25880000 25860000 25840000 0 10 20 OBA 30 Time O_UE FW 40 Path Based 50 60 ©Association for European Transport and contributors 2006 .FIGURE 2: OBJECTIVE FUNCTION vs.Chicago Net..

01 UE Path-Based Bar-Gera OBA Origin UE 618. Network.7 16. Gap– Chicago Reg.09 0.01 UE Path-Based Bar-Gera OBA Origin UE 4.025 0.5 NA 9.15 .7 572. Network.6 81.01 0.002 0.9 .0001 0.14 0. 2GHz Pentium M . Network.021 .0025 0.016 0.01 UE Path-Based Bar-Gera OBA Origin UE 0.0 9.0 11.4 19.9 .9 TABLE 3: Average excess costs at Rel.001 0.001 15.0001 93.TABLE 1: CPU Minutes to a Relative Gap – Chicago Reg.0 175.8 .8 TABLE 2: Max flow change at Relative Gap– Chicago Reg.0001 5. 2GHz Pentium M .7 34.0 5.9 112.0 26.7 NA 81.002 ©Association for European Transport and contributors 2006 .6 NA 248.1 66. 2GHz Pentium M .4 0.3 .001 40.23 0.

0025 n/a 0.01 UE Path-Based Bar-Gera OBA Origin UE 352.2512 0.0001 1:05:25 2:08:08 2:14:26 0:21:37 TABLE 5: Max flow change at Rel.2GHz Dual Athlon .03 n/a 50.0001 3.2541 0.0001 0.2GHz Dual Athlon . excess costs at rel. Network.01 UE Path-Based Bar-Gera OBA Origin UE 0.001 55.59 n/a 81.001 0:11:32 0:32:21 0:52:05 0:10:46 .89 23.2GHz Dual Athlon .017 0.0231 . 2. gap– Chicago Reg.22 n/a 119. 2.68 . Network.001 0.TABLE 4: CPU Time (Hours/Minutes/Seconds) to a Rel.01 UE Path-Based Bar-Gera OBA Origin UE 0:2:43 0:08:43 0:12:37 0:04:33 .021 0. Gap – Chicago Reg.84 TABLE 6: Avg.2153 .196 0. Network.74 62. Gap– Chicago Reg.89 .2566 0. 2.0024 ©Association for European Transport and contributors 2006 .44 579.0025 0.

0259 0. Gap – Philadelphia.78 n/a 763 .01 UE Path-Based Bar-Gera OBA Origin UE 3596.02 TABLE 9 Average excess costs at relative gap– Philadelphia.88 n/a 139.2173 0.048 0.0022 ©Association for European Transport and contributors 2006 .01 UE Path-Based Bar-Gera OBA Origin UE 0.0073 0. 2.49 8327.001 630.0001 0:27:05 0:12:14 0:27:06 0:12:45 TABLE 8: Max flow change at relative gap– Philadelphia.01 UE Path-Based Bar-Gera OBA Origin UE 0:01:26 0:03:24 0:10:51 0:03:05 .2 GHz Dual Athlon .00231 0.0224 .0001 85. 2.0001 0. 2.114 0.001 0.022 0.11 .2 GHz Dual Athlon .001 0:05:01 0:08:30 0:17:24 0:05:27 .TABLE 7: CPU Time (Hours/Minutes/Seconds) to a Rel.00226 0.10 1857.0 .1178 0.75 n/a 1873.57 1566.2 GHz Dual Athlon .

Regional Network.01 UE Path-Based Origin UE 1554.C. Regional Network.001 0:19:17 2:44:45 0:23:29 .001 0.1853 0.35 TABLE 12: Avg.1414 0.96 3725.TABLE 10: CPU Time (Hours/Minutes/Seconds) to a Rel. 2.01 UE Path-Based Origin UE 0:04:41 0:48:56 0:11:21 .64 .0175 0.00177 ©Association for European Transport and contributors 2006 .12 634.63 437. Excess costs at rel.69 35. Gap – D. gap.0001 0.01 UE Path-Based Origin UE 0.C. Regional Network.0018 0.D.0001 47.2 GHz Dual Athlon .D. 2.2 GHz Dual Athlon .89 1265.001 190. gap.0001 1:21:40 5:22:28 0:42:06 TABLE 11: Max flow change at rel.1290 .0099 0.12 .2 GHz Dual Athlon .C. 2.0148 .0016 0.70 194.

0035 0.0001 0.0045 0. Regional Network 0.0010 0.15 2.38 171.001 0.000001 Relative Gap– Washington.0005 0.97 2143.769 Max Flow Difference 2719.0015 0.0050 0.935 10.FIGURE 3: Rates of Convergence for One and Two Core ProcessorsWashington. D.0040 0.68 ©Association for European Transport and contributors 2006 .0020 0.214 39.0025 0.00001 RMSE 111.C. D.85 724.01 0. Regional Network GAP 0.0000 0:00:00 0:15:00 0:30:00 0:45:00 Time 1:00:00 1:15:00 1:30:00 TABLE 13: COMPARISON OF UE-FW Link Flows to OUE Link Flows Computed to a 0.C.0030 HP1 HP2 GAP 0.

08 0.TABLE 14: Comparison of Warm and Cold Start Assignments on the Washington DC Regional Network Time to .000090 9 0.37 39.001718 44935738.0001 RG Original 00:41 Assignment Increase 00:12:10 Rand 5% w Warm Start 00:49:25 Increase Rand 5% without warm start # Iter. Relative Gap Max Flow Delta 0.001673 42477153.00 146.000089 39 0.001854 44936192.000096 ©Association for European Transport and contributors 2006 .36 0.17 23.10 AEC Objective Function 32 0.