Bus Timetables with Even Passenger Loads as Opposed to Even Headways
Transportation Research Institute, Civil Engineering Faculty, Technion-Israel Institute of Technology. Haifa, Israel 32000
Abstract. Bus Timetables with even headways often expect the passengers to adjust themselves to given timetables rather than adjusting the timetables to the fluctuated passenger demand. This expectation is one of the largest sources of unreliable service which extends passenger waiting times at the bus stops, and create overcrowdings. This work attempts to create public transport timetables so as to improve the correspondence of vehicle departure times with passenger demand. The methods presented can be carried out in an automated manner. With the growing problems of transit reliability, and advance in the technology of passenger information systems, the importance of even and clock headways is reduced. This allows for the possibility to create more efficient schedules from both the passenger and operator perspectives. The methodology framework contains a developed algorithm for the derivation of vehicle departure times (timetable) with even average loads and smoothing consideration in the transition between time periods. The procedures presented are accompanied by examples and clear graphical explanations. It is emphasized that the public timetable is one of the predominant bridges between the operator (and community) and the passengers.
There is a saying that “A stitch in time would have confused Einstein”. Along this line one can say that many stitches in a public transport timetables would confuse the passengers. No doubt that the public timetable is one of the predominant bridges between the operator (and/or the community) and the passengers. Therefore more attention should be provided for the construction of timetable in order to improve its correspondence with the fluctuated passenger demand. In general terms, the public transport operational planning process includes four basic components performed in sequence: (1) network route design, (2) setting timetables, (3) scheduling vehicles to trips, and (4) assignment of drivers (crew). It is desirable for all the four components to be planned simultaneously to exploit the system’s capability to the greatest extent and maximize the system’s productivity and efficiency. However this planning process is extremely cumbersome and complex, and therefore seems to require separate treatment of each component, with the outcome of one fed as an input to the next component. In the last twenty years, a considerable amount of effort has been invested in the computerization of the four components mentioned above, in order to provide more efficient controllable and responsive schedules. The best summary as well as the accumulative knowledge of this effort was presented in the second through the seventh International Conferences on Transit Scheduling, and appear in the book edited by Wren (1981), Rousseau (1985), Daduna and Wren (1988), Desrochers and Rousseau (1992), Daduna, Branco, and Paixao (1995), and Wilson (1999). Mathematical programming methods for determining frequencies and timetables have been proposed by Furth and Wilson (1981), Koutsopoulos, Odoni, and Wilson (1985), Ceder and Stern (1984), and Ceder and Tal (1999). The objective in Furth and Wilson is to maximize the net social benefit, consisting of ridership benefit and wait time saving, subject to constraints on total subsidy, fleet size and passenger loading levels. Koutsopoulos et al extended this formulation by incorporating crowding discomfort costs in the objective function and treating the time dependent character of transit demand and performance. Their initial problem comprises a non-linear optimization program relaxed by linear approximations. Ceder and Stern addressed the problem with an integer programming formulation and
These methods aimed at practicability appear in Ceder (1984. In section 3 the scope and framework of this study are outlined. The basic criteria for the determination of frequencies are: (a) to provide adequate vehicle’s space to meet passenger demand. and advance in the technology of passenger information systems. In the ride-check methods the frequency is the division between and
. In the point-check methods the frequency is the division between passenger load at the maximum (max) load point (either the one across the day or in each hour) and the desired occupancy or load factor. and (b) to assure a minimum frequency (maximum-policy headway) of service. Ceder (1984) described four
different methods for calculating the frequencies. Two are based on point-check (counting the passengers on-board the transit vehicle at certain point(s)). the importance of even headways and clock headways (see definitions next section) is reduced. In section 5 the proposed algorithm is interpreted and implemented graphically. The latter approach focuses on reconstructing timetables when the available vehicle fleet is restricted. 1986) and are briefly described in the following (Background) section 2.
2. Finally Ceder and Tal used mixed integer programming and heuristic procedures for constructing timetables with maximum synchronization.
Public transport timetable is commonly constructed for given sets of derived frequencies. Other methods for frequency and timetable determination are related to the type and adequacy of the input passenger count data. Sections 6 and 7 provide an example and concluding remarks. This allows for the possibility to create more efficient schedules from both the passenger and the operator perspectives.heuristic person-computer interactive procedure. and two – on ride-check (counting the passengers along the entire transit route). Finally is worth mentioning that with the growing problems of transit reliability. That is maximization of the number of simultaneous arrivals of vehicles to connection stops. In section 4 an algorithm is proposed for the derivation of vehicle departure times with even average loads and smoothing techniques in the transition between time periods.
This analysis allows for establishing a spectrum of alternative timetables.
3.average or restricted-average passenger load and the desired occupancy. Each timetable is accompanied by two comparison measure which are used as an evaluation indicator in conjunction with resource saving. and (c) selection of special requests. In category (a) the headway (time interval between adjacent
departures) can be equal or balanced. in order to assure that in certain percentage of the route length the load does not exceed the desired occupancy. and the restricted average is a higher value than the average one. These cases are being extended in this work. In category (b) it is possible to select for each time period one of the four frequency determination methods (two point-check. The first measure is the total required vehicle runs (departures) and the second is an estimate for the minimum required fleet size at the route level only. number of seats).g. easy-to-memorize) and/or certain number of departures (usually for cases with limited resources). using passenger load data. or a given frequency by the scheduler. Equal headway refers to the case of evenly spaced headways and balanced headway – to the case of unevenly spaced headways but with even average passenger load at the hourly maximum load point. (b) selection of frequency determination method for each period. In category (c) it is possible to request clock headways (departure times that repeat themselves in each hour. This desired occupancy (or load factor) is the desired level of passenger load on each vehicle. In a follow-up study Ceder (1986) analyzed optional ways for generating public timetables. based on three categories of options: (a) selection of type of headway. The average load is determined by the area under the load profile (in passenger-km) divided by the route length (km). and two ride check) mentioned above. The outcome of these analyses is a set of optional timetables in terms of vehicle’s departure times at all specified timepoints. The assumption that passengers will adjust themselves to given timetables
. in each time period (e. Scope
Transit public timetable is perhaps the main reference for defining unreliable transit service.
say. 1 illustrates the research progress regarding the construction of public timetables. Both situations are not observed when the service is highly frequent and characterized by low variance of the headway distribution. All the Timetables (1.2. Fig. where the average passenger loads are counted at the max load point of the route for each vehicle separately. Level 1 is related to the studies of Ceder (1984. Level 2 is related to the study of Ceder (1986). Level 3 represents this work where the average passenger loads are counted at each vehicle’s max load point as opposed to the route max load point at level 2. and the desired occupancy is kept in these transition periods. This algorithm is then applied graphically in section 5. longer that 10 minutes) instead of adjusting the timetables to the passenger demand is one of the largest sources of unreliable service.3) in Fig. and the division of this load by the desired occupancy results in the frequency unless the minimum required frequency is not reached. This will eventually lead to the known bunching phenomenon with the vehicles behind. This will result in a more reliable and comfortable service. It is the purpose of this work to establish a method for better matching the passenger demand with a given timetable while attempting to minimize the fleet size (one of the main resources). where the average passenger load is counted at the max load point on an hourly basis.(with headways of. When passenger demand is not met.1 are also based on a smoothing procedure between the time periods such that there is no need to round any number. the departure times (Timetable 1) at the hourly max load point are determined with even headways. In order to derive the departure times (Timetable 3) with even load at the critical max load point of each vehicle an algorithm is developed in the next section. These loads are accumulated with respect to time and based on the desired occupancy values the departure times (Timetable 2) at the route maximum load point are determined with uneven headways and even average loads only at this route point. 1986). Opposite to that is the situation of overestimating the demand which may result in transit vehicles running ahead of time. the transit vehicles are slowing down (increased dwell time).
. The non-integer value of the frequency is then kept and based on the accumulative frequency curve (adding the frequency at each hour with respect to time). behind the schedule and entering the inevitable process of further slow down.
1 Optional Timetables from a Balanced Passenger Load Perspective
. Even Loads at each Vehicle’s Max Load Point
Passenger Loads on Individual Vehicles at the Hourly Max Load Point
Uneven Headways.Level 3
Passenger Loads on Individual Vehicles at each Max Load Point
Uneven Headways. Even Loads at the Hourly Max Load Point
Passenger Loads at the Hourly Max Load Point
v. 3. 2. represents uniform arrival rate of passengers. 2.1 Definitions and Objective
Let: Li ( t ) = accumulative load (# of passengers) curve at stop i w. 2. 2. u = 0. which is the basic input of the analysis and has the same slope (straight line) between each two adjacent departures. i = 1. n. …. 2. ….4. …. 1. …. v. …. i = 1. = transition time between time intervals.
= the average service travel time between the departure terminal i = 1 and stop i . v.
j = 1. and the last slope is extrapolated to the schedule (timetable) horizon. n ( T1u = 0) during time interval u = 1. u = 1. u=1. m. where changes in the slope are made at the departure times of the transit vehicles at i . …. v. …. An Algorithm for Balancing the Critical Loads
Fmu = minimum required frequency (# of vehicles) in interval u. That is. The objective set forth is to change the departure times such that all observed average max loads will be same and equal to d u during all u .
The purpose of the algorithm presented below is to derive the transit timetable provided that in an average sense all vehicles will have even load (equal to the desired occupancy) at the max load stops of each vehicle. ….th candidate departure time at stop i . time t.t. for a given time period each vehicle may have a different max load point across the entire transit route with a different observed average load.r. i = 1 2. u and u+1. ….
u = 1. The curve Li ( t ) . = desired occupancy (load factor) on each vehicle for each time interval u . 2. n. 2.
= the derived j .
d um = desired occupancy for situations when Fmu is applied d um < d u . 2. Nu
= number of derived departures for t u −1 ≤ t < t u .
…. j +1 by calculating Li t i . Proof and Explanation The basic assumptions of Theorem 1 are: (a ) the change of departure times will not affect the arrival pattern of passengers. 2. j +1 crosses the schedule (timetable) horizon STOP. j +1
)] ) = L (t
+ Tiu + d u . where t i . and (b) the change of departure times (with same frequency) will not affect the passenger demand. Since the max load point is commonly defined for a time period of one hour or more. 2 . n Step 1 : Determine the j − th departure time (during time interval u ) at t ∗1 j such that t ∗1 j = min t 1 j .
Theorem 1 Algorithm T produces departure times for individual transit vehicles such that their maximum average load equals d u . t nj − Tnu Step 2 : Determine ti . ….
.. j +1 belongs to u .
Note: This algorithm also incorporates the criterion of minimum frequency through a certain check and adjustment described below in this section. When t i . 10 minutes. for all
i = 1. Otherwise set j + 1: = j and go to Step 1..Certainly the adjustments in the timetable are not intended for highly frequent urban services where the headway is less than say. Behind this algorithm is the notion that passenger overcrowding situations (loads greater than d u ) should be avoided. there are situations in which the critical points for individual vehicles do not coincide with this max load point. n. on an hourly frequency of about 6 vehicles or more..
4. t 2 j − T2 u .2
Initialization : Set j = u = 1 and determine ti1 by calculating L(t i 1 ) = d 1 for all i = 1 . These critical points (max load points for individual vehicles) are treated by the following Algorithm T..
with a straight line (on this curve) between each two adjacent departures. time... Each increase of
Lq (t ) by du is coordinated with a time point to be the resultant departure time (Ceder. That is.
4. This is the case for a single max load point in a given time period (usually one hour). whenever a new t1∗j becomes ahead of the next t u . in the transition between time intervals. then Lq (t ) is the curve of accumulated max loads (at q) vs.Given these two assumptions. rather than by Li ( t ) as in Algorithm T. for each time interval u. i ≠ q .
1986). if the
departure time of a trip will be set at t qj − Tqu (at the route dep. then tij − Tiu ≤ tqj − Tqu . will be met. and hence may create overcroadwing at the critical point i. If t ij will be derived by Lq ( t ) . Tqj − Tqu . the new (balanced load) departure times are constructed on an accumulative load curve.. However. and that the max average load at any stop will be d u . Fmu is checked and attained in the following manner:
.. Algorithm T refers to the case of different max load points. by coordinating the appropriate cumulative desired occupancy value. Usually Fmu is needed in the beginning and end of the day. the average observed load at i will be equal or greater than d u (and exactly d u at q).. i.. That is.3
Minimum Frequency Criterion
Algorithm T or the balanced load procedure does not guarantee that the minimum (min) frequency criterion (inverse of the policy headway). tij − Tiu . d u .. during the process of Algorithm T..e. Therefore. Since Li ( t ) and
Lq ( t ) are two accumulative passenger load curves vs. Fmu ..
In others words. Fmu needs to be checked. time. Considering N u which is the number of derived departures by Algorithm T during
tu −1 ≤ t < tu . the fact that tij − Tiu is less
than or equal to t qj − Tqu means that Li t qj ≥ Lq t qj . in the case that stop i comes before stop q (similarly is the other case when stop q comes before i). and the time axis. point). Algorithm T assures that this cannot be the case.. according to the algorithm t1∗j = tij − Tiu = min .. if i = q is the hourly max load point. each is associated with a different transit vehicle.
set tij = ti . Both cases can be observed by the accumulative load curves or load profiles. (Three)change d u to dum in Step 2 of Algorithm T and continue until t1∗j ≥ t u . where in the
Initialization d1 = d um . dum . N u ≥ Fmu ?. There are two more notes worth mentioning about Algorithm T and the min frequency case: (1) if for a given departure to be determined.…. otherwise – END.2..
5. a different type of vehicle (than the one for which d u is set) is considered d u can be changed in Step 2 of Algorithm T. if yes –END. for the min frequency situation:
i = 1. 2 . (Two) calculate the new desired occupancy. in (c). if it is preferred to use a policy headway criterion (inverse of Fmu . The procedure described by (a). Fig.n. This dum
replaces then d u . i=1. ….(One) check for each u. 2. 2 illustrates a simple example of the algorithm.. j −1 + and Fmu Fmu continue with Algorithm T. in a transit line with i=1 as the departure point and i=n+1 as the arrival point. for each u) then in the procedure for Fmu . for each derived departure time (starting 1 1 with t ’ij ) there will be an extra check: if tij − ti . j −1 > .
. Graphical Interpretation of Algorithm T
The principal of Algorithm T can be demonstrated using the accumulative passenger load curve at each stop i. If Algorithm T results in time interval u with a
frequency less than Fmu then the maximum difference in passenger loads on
Li ( t ) between tu and tu − 1 determined a new desired occupancy. otherwise continue u = 1. This may be the case for excessive load which may result in a too short headway or large amount of empty seat-km resulting in Fmu .. then END. n
max Li ( tu ) − Li ( tu − 1 ) Fmu + 1
and return to Algorithm T at tu −1 .. v. until the derivation of the first departure time in interval u + 1. (b) and (c) assures that the min frequency criterion will always be met. dum .
T ra ve l T im e:
O b serve d D ep artu res: D esired O cc u p a nc y :
6:15 .R o u te :
1 5 m in .
6 :4 5.
50 p a sse n ge rs
Ac cu m u la tive P a sse n g e r L o a d
200 15 0
7 :10 -Departure
STO P A
0 6:00 30 6:2 2.5 6 :31 6:58
200 18 0 150
ST OP B
6 :45 6:56
Fig 2 Interpretation of Algorithm T
6:45.Given a transit line A B C with average travel time of 15 minutes between A and B. 6:45 and 7:10.5 at A. and on the 7:10 vehicle: 25 and 80 passengers at A and B. 2.between B – C. Step 2 then selects 6:31 as the next departure. 100 and 150 with the time axis. and 180 . It means that the first vehicle will be shifted backward by 4 minutes to have at B. and 65 at stop B. with about 17 minutes even headways. 2 shows LA ( t ) and LB ( t − 15) as the accumulative load curves of the three vehicles where the curve at B is shifted by 15 minutes to allow for an equal time basis (at the route’s departure point) in the analysis. 1 for comparison purposes. and by coordinating the desired occupancy value of 50. 6:00 to 7:10. On the 6:45 vehicle: 80 and 35 passengers at A and B. and the algorithm continues and results in 6:56 as the last departure at the period [6:00 . Fig. It means that the max load point for the vehicles is point B. Algorithm T continues in Step 2 by adding d u = 50 to 50 at LB ( t − 15) . 7:01. Coming back to Algorithm T (Level 3 in Fig. 3 is summarized in Table 1 below where the associated average max load its corresponding stop appear in brackets. one obtains 3. The average observed on-board loads on the 6:15 vehicle are 30
passengers at stop A. say. The load profile of these 3 vehicles is the histogram of 135 passengers between A – B. and a desired occupancy of 50 passengers. respectively. For Level 2 LB ( t − 15) in Fig. The comparison between the observed data and the results of Levels 1.
. This simple example can be subjected to Level 1 and Level 2 analyses in Fig. 3). 3 is the necessary curve. Adding d u = 50 to 122 (at A) or to 134 (at B) will results in departures beyond 7:10. one obtains the departure at 6:11. 1) then at the initialization the value of 50 is coordinated with LA ( t ) and LB ( t − 15) to obtain: 6:11 at B and 6:22. This is for Level 1. with three departures at 6:15. Step 1 of the Algorithm T selects the minimum time between the two to be the first departure at 6:11 (emphasized in Fig. in an average sense. Dividing 180 by the desired occupancy 50. This results in t A 2 = 6:31 and t B2 = 6:45 (for j=2). 50 instead of 65 passengers. and to L A ( 611 : ) =22 at LA ( t ) .6 vehicles during the considered time period. In fact this shift in the time scale complies with ( tij − Tiu ) in step 1 of Algorithm T.7:10].
6:11 Level 3
(50. At the 6:17 dep. (46. 2. of the 6:34 dep. at 7:08 with max load of 74 at B. 3). A)
Even Headway Uneven Headway. B)
* average max load on each vehicle at its specified stop **Level 1 results also in a 4th dep. intersects LB ( t − 15) at 87. Individual Max Load Point
6:11 Level 2
For example the max load at dep. B)
(50.20)=46 at A. is max. B)
(80. Fig. and results in 46=81-35 passengers. A)
(51. Then the 6:34 dep. and 35 passengers at A (intersect between 6:17 and LA ( t ) in Fig. The max load. Combined Max Load Point Uneven Headway. 6:34 (Level 1) is determined from the difference in average loads between the 6:34 and the 6:17 departures. A)
(50. from this example. and results in 20=87-67 passengers at B. 3 (1st example)
(65. the combined max load point of the three vehicles. the sum of the max loads is neither same for each Level nor for the observed data. Departure times associated with Levels 1. Similarly the 6:34 dep. 2. Another example for the interpretation of Algorithm T can be based on the example appeared in Ceder (1986). intersects LA ( t ) at 81.Table 1. B)*
(46. the initial accumulate passenger load curve at the hourly max load points (Level 2). 67 passengers will be observed at stop B based on LB ( t − 15) in Fig. B)
6:17 Level 1
(67. therefore. 3 exhibits. Algorithm T is then applied for this example given the loading levels of each of the four vehicles
. Consequently the balanced load along the entire route is attained only at Level 3 (Algorithm T) whereas at Level 2 this balanced is attained only at stop B. A)
(35. Certainly due to the different derived departure times for the 3rd departure (much below the observed 7:10).
That is. 3. 1)
(71.2. 4 in which two departures are associated with the time interval 6-7 a.at each stop along the entire 10-km transit route. 1)
(72. 1. 63 passengers and those of 4 more departures not in the example). 27+50. 2. The load profile of each of the four vehicles (dep. and 3 is shown in Table 2:
Table 2. 2.m.
. the hourly max load point between 6-7a. 1. This interpretation of Algorithm T is shown in Fig. Combined Max Load Point Uneven Headway. Departure times associated with Levels 1. Then du = 50 is added to Li ( t − Ti1 ) . 1)
(56. i=Dep. 18 respectively. **with extrapolation over 7:15 at the Dep Stop. the comparison between the observed four departures and the results of the analyses at Levels 1. 1. 1. 1)
(65. The minimum second departure is then determined at 6:46. (the max loads observed are 56.m. and stop 2 between 7-8a.Dep)
6:23 Level 1
(32. 43+50. Algorithm T starts with du = 50 to determine the minimum dep. 2. Individual Max Load Point
6:33 Level 2
(70. and 3. 3 (2nd example)
(35. 3. 2)
(50. Similarly to the previous example. 1)
Even Headway Uneven Headway. 3).Dep)
6:31 Level 3
(63. On the left hand side there are four accumulation curves associated with 4 stops: Dep. When applying Level 2 analysis (Fig. 2)
(72. and the other two at 7:05 and 7:14. is stop 1 with 23+67=90 passengers. and two (out of six that appear in Ceder (1986)) between 7-8 a. 4 with an emphasize on each individual vehicle max load.
Ti1 = 0.m.Dep)
* average max load on each vehicle at its specified Stop (Dep.) along the 10-km route is illustrated on the right hand side of Fig. 2)*
respectively. and 43+50 for i=Dep.1). 7. time associated with d u at 6:17 (stop 2).
6 vehicles during the considered time period. 6:45. and to L A ( 611 : ) =22 at LA ( t ) . Algorithm T continues in Step 2 by adding d u = 50 to 50 at LB ( t − 15) . and by coordinating the desired occupancy value of 50.between B – C.
. 6:45 and 7:10.7:10]. 6:00 to 7:10. respectively. one obtains 3. and a desired occupancy of 50 passengers. Dividing 180 by the desired occupancy 50. Coming back to Algorithm T (Level 3 in Fig. and 65 at stop B. This results in t A 2 = 6:31 and t B2 = 6:45 (for j=2). 3 is summarized in Table 1 below where the associated average max load its corresponding stop appear in brackets. Fig. This is for Level 1. with three departures at 6:15. The load profile of these 3 vehicles is the histogram of 135 passengers between A – B. one obtains the departure at 6:11. The comparison between the observed data and the results of Levels 1. It means that the max load point for the vehicles is point B. This simple example can be subjected to Level 1 and Level 2 analyses in Fig. Adding d u = 50 to 122 (at A) or to 134 (at B) will results in departures beyond 7:10. with about 17 minutes even headways. 2. 1) then at the initialization the value of 50 is coordinated with LA ( t ) and LB ( t − 15) to obtain: 6:11 at B and 6:22. in an average sense. and on the 7:10 vehicle: 25 and 80 passengers at A and B. and 180 .5 at A.Given a transit line A B C with average travel time of 15 minutes between A and B. 3). and the algorithm continues and results in 6:56 as the last departure at the period [6:00 . 7:01. Step 1 of the Algorithm T selects the minimum time between the two to be the first departure at 6:11 (emphasized in Fig. In fact this shift in the time scale complies with ( tij − Tiu ) in step 1 of Algorithm T. 100 and 150 with the time axis. 1 for comparison purposes. Step 2 then selects 6:31 as the next departure. On the 6:45 vehicle: 80 and 35 passengers at A and B. It means that the first vehicle will be shifted backward by 4 minutes to have at B. 2 shows LA ( t ) and LB ( t − 15) as the accumulative load curves of the three vehicles where the curve at B is shifted by 15 minutes to allow for an equal time basis (at the route’s departure point) in the analysis. 50 instead of 65 passengers. 3 is the necessary curve. say. The average observed on-board loads on the 6:15 vehicle are 30
passengers at stop A. For Level 2 LB ( t − 15) in Fig.
therefore. Consequently the balanced load along the entire route is attained only at Level 3 (Algorithm T) whereas at Level 2 this balanced is attained only at stop B. Another example for the interpretation of Algorithm T can be based on the example appeared in Ceder (1986). the combined max load point of the three vehicles. Then the 6:34 dep. B)
(50. Similarly the 6:34 dep. Algorithm T is then applied for this example given the loading levels of each of the four vehicles
6:17 Level 1
(67. Combined Max Load Point Uneven Headway. intersects LA ( t ) at 81.Table 1. B)
(80. Individual Max Load Point
6:11 Level 2
6:11 Level 3
(50. 3 exhibits. at 7:08 with max load of 74 at B. and 35 passengers at A (intersect between 6:17 and LA ( t ) in Fig. At the 6:17 dep. the sum of the max loads is neither same for each Level nor for the observed data. 3 (1st example)
* average max load on each vehicle at its specified stop **Level 1 results also in a 4th dep. intersects LB ( t − 15) at 87. The max load. A)
(35. 2. A)
(50. Departure times associated with Levels 1. from this example. of the 6:34 dep. 6:34 (Level 1) is determined from the difference in average loads between the 6:34 and the 6:17 departures. 67 passengers will be observed at stop B based on LB ( t − 15) in Fig. the initial accumulate passenger load curve at the hourly max load points (Level 2). 2.20)=46 at A.B)
Even Headway Uneven Headway. B)*
(46. Fig. is max. Certainly due to the different derived departure times for the 3rd departure (much below the observed 7:10). (46.
For example the max load at dep. 3). and results in 20=87-67 passengers at B. and results in 46=81-35 passengers.
That is. 1)
Even Headway Uneven Headway. Combined Max Load Point Uneven Headway. is stop 1 with 23+67=90 passengers. Algorithm T starts with du = 50 to determine the minimum dep.
Ti1 = 0. Then du = 50 is added to Li ( t − Ti1 ) . 1)
(63. 2. The minimum second departure is then determined at 6:46.m.Dep)
* average max load on each vehicle at its specified Stop (Dep. 1)
(70.at each stop along the entire 10-km transit route. 43+50. 1. 1)
6:31 Level 3
(54. This interpretation of Algorithm T is shown in Fig. 3. and the other two at 7:05 and 7:14. 3). 1)
(72. 2. **with extrapolation over 7:15 at the Dep Stop.m. 1)
(71. Similarly to the previous example. 1.1). 27+50. 4 with an emphasize on each individual vehicle max load. 2)
(50. and stop 2 between 7-8a. 1)
(65.m.) along the 10-km route is illustrated on the right hand side of Fig. 3 (2nd example)
(56. The load profile of each of the four vehicles (dep. 2. i=Dep. 1. (the max loads observed are 56.
. 1. and 3. and two (out of six that appear in Ceder (1986)) between 7-8 a. 18 respectively. On the left hand side there are four accumulation curves associated with 4 stops: Dep.m.
respectively. Individual Max Load Point
6:33 Level 2
(55. 3. Departure times associated with Levels 1. 2)*
(67. When applying Level 2 analysis (Fig. and 3 is shown in Table 2:
Table 2. the hourly max load point between 6-7a. 4 in which two departures are associated with the time interval 6-7 a. and 43+50 for i=Dep. 63 passengers and those of 4 more departures not in the example). the comparison between the observed four departures and the results of the analyses at Levels 1. time associated with d u at 6:17 (stop 2).Dep)
6:23 Level 1
Accumulative Passenger load
50 0 6:00
:10 :20 :40 7:00 :40
Fig. 3 Accumulative Passenger Load Curve using Level 2 Analysis
Å Stop 3
7 min. (6:45)
(t . Stop and each Stop (Dep.8 a.
* Includes 4 more departures between 7. 1 2 3 Arrival Stop #
60 40 20
98 48 18
Accum ulative Passenger Load
6:25 6:45 7:05 7:15 200 Dep. 2
Å Stop 1
Dep. to Dep. 6:45 7:05 7:15 23 67 56 63
Max.m. Load Point Stop 1 Stop 2*
Average Travel Time between
43 50 0
(t . Load re Point Time 6:25 a.7 )
6:46 7:05 204
100 50 0
Veh. (km )
at the Hourly Departu Max. 12 min. 4 Interpretation of Algorithm T for the example in Fig. (7:05)
(t . (6:25)
7:00 7:14 210
50 42 30
Dep. 18 min.12 )
60 40 20
Distance from Dep.
.m. 93-21. However the loads at 6:46 are being accumulative.m.. for 6-7 and 7-8 a. Finally the departures at Level 3 are carried out by Algorithm T with the initialization phase shown on the four accumulative curves in Fig. 93. the max load is 32 at i=2. hence. 6-7 a.m. appears underneath the dep. 3 attains 132 at 7:00. respectively. Once again only at the Level 3 analysis the average max loads attains the desired occupancy at each time interval ( d u = 50 . The calculations of the max loads are done in the same manner as for Level 1. Table 3 contains the necessary information and data for a 3-hour example of a transit line from A to B and B to A. and the loads at 6:46 are 51. is based on 4 more departures. Point B can be perceived as the CBD that attracts the majority of the demand between 6-9 a. The departures at Level 2 are based on the accumulative curve in Fig. For example at 6:23 the loads on Li ( t ) are 17. the accumulative load curve in Fig.
6. 7-8 a. 88-32. 4. 3. 2. and
d u = 65 . 28 passengers for i=Dep. 74-28)=72 at stop i=1 . respectively. There are 14 and 8 departures for A to B and B to A. 32.
.The departures at Level 1 are based on even headways of 23. 88. and 9 minutes. The loads at 6:23 are associated with individual vehicles and.) whereas at the other analyses the undesirable imbalanced load
exists at the critical points.m. The average observed max load on each trip is also shown in Table 3.m.. 1.64 (veh/hr) with the inverse of 23 minutes. 3. Example
The examples in this works are used as an explanatory device for the developed procedures. 46 for the four stops. time.. respectively. is calculated from the previous dep. The 9 minutes headway between 7-8 a. 21. and dividing this load by d u = 50 loads to the frequency of 2. That is. with extrapolation to 230 (2x50+2x65) for the 4th dep. The average max load. end hence the max load is max (51-17.
and the third at level of 60+50=110 passengers. which doesn’t comply with Fm =2 . is applied. The third dep. Therefore d u is changed to 65 and the third dep. and B-A directions) and Tm1 is the average travel time from A to m. in Fig. whereas Fm =2 in Table 3. 5 and 6. u=2. these Table 3 data are used for running Algorithm T.
departure between 6-7 a. Step (b) of this complementary component provides the formula for the new desired occupancy: d1m =
L p (7:00) 2 +1 = 91 .m. There are 7 new departures for direction B to A in Fig. 6. 6 is related to the policy headways. that are based on desired occupancies of 50 and 65 passengers. That is.m. That is. is coordinated with L p (t −Tm1 )=165 to be 7:07 a. Algorithm T determines the new departure times shown in Figs. at first d u =1 =50 is set and results in one departure between 6-7 a.). Another observation in Fig. 5. where p is the single max load point (both for A-B. Since Fm =2 is the only requirement.In order to construct the balanced load timetable. There are 14 new departures for direction A to B. if there is a policy (max) headway criterion then the note beneath the description of the minimum frequency component of Algorithm T (section 4.m.). Assuming that the max load is observed at the same stop for each direction. in Fig.m. 3 (7-9 a.m.
. d u =50 .). 6. the second dep. results in a departure over 7:00 a. (7:04 a. For the B-A direction.m. For the A-B direction. the first two departures are at load levels 3
of 30 and 60. there is a large headway between 7:44 and 8:32 departures. However..m. 7.m. u=1(6-7 a. check for L p (t −Tm1 )=150 . and hence the minimum frequency complementary component of Algorithm T. at
’ Consequently there is only one L p t −Tm 1 =100 is beyond 7:00 a. in Fig.2) can be used. and d u =65 .
m.Table 3. 6:20 6:40 6:50 7:05 7:15 7:25 7:30 7:40 7:50 8:00 8:10 8:20 8:35 8:50
B A 6:30 6:45
52 43 65 59 50
8-9 a. Point Time
Average Observed Max Number of Passengers on board the vehicles A B 15 30 47 58 65 79 90 82 62
Desired Occupancy (Pass.
for all hours both directions.)
A B 6-7 a. Given Data for the Example Problem
Departure Time at the Route Dep.
8:25 8:40 8:55
68 55 80 71
23 51 28
Minimum Frequency : 2 Vehicles per hour.m.
. 6:52. 5 Determination of Balanced Load Departure Times for the Example Problem. 8:25. 7:54. 7:35. 8:37.Accum ulative Passenger Load
726 700 671
528 500 466
215 200 150 65 100 50 50 15 0 6:00 :20 :40 7:00 :20 :40 8:00 :20 :40 9:00 45 92
Dep. 7:07. 8:13. 8:51
Fig. 8:03. 7:44. Tim e : (determined) 6:41. 7:29. 7:25. direction A-B. 7:17.
7:44. there is a need when
developing computerized procedures to supply the schedulers with alternative schedule options along with interpretation and explanation of each alternative. 6:45. 8:52
Fig. 7:26. Discussion and Concluding Remarks
Different public transport agencies use different scheduling strategies based primarily on their own schedulers’ experience.Accum ulative Passenger Load
316 300 237 214 200 155 50
50 0 6:00
22 :20 :40 7:00 :20 :40 8:00 :20 :40 9:00
Dep. 7:09. at the detailed level. One
. direction B-A. 8:32. Tim e : (determ ined) 6:33. Consequently. In addition.
7. even at the same public transport agency. 6
Determination of Balanced Load Departure Times for the Example Problem. the schedulers may use different scheduling procedures for different groups of routes. it is unlikely that two independent public transport agencies will use exactly the same scheduling procedures. As the result. and secondarily on their scheduling software (if any).
This allows for introducing optional timetables with the
consideration of even average loads on individual vehicles.such alternative is developed in this work. With the growing problems of public transport reliability. Theophrastus (300 B. The controlled procedures for adjusting the timetable. it is desirable that one of the alternatives will coincide with the scheduler manual procedure. social and recreational public transport needs of the community. In cases of large empty-seat (space)km the desired occupancy can be lifted while controlling. industrial. made in this work by Algorithm T. The adjustments of departure times. It is known that passenger demand varies even within one hour. The keyword here is to be able to control the loading instead of being exposed repeatedly to an unreliable service resulted from imbalance loading situations. These adjustments resulting in a balanced load
timetables are based on a given vehicle desired occupancy at the maximum load point of each vehicle. Also. The construction of such timetables takes into account. In this way. in an average sense. reflecting the business. the scheduler will be in a position not only to expedite manual tasks but also to compare methods with others regarding the trade-off between passenger comfort and operating cost. in essence. form the basis to improve the correspondence of vehicle departure times with the fluctuated passenger demand. and advance in the technology of passenger information system the importance of even and clock headways is reduced. and attempts must be made to avoid that passengers will spend unnecessary time when using public transport. Average even loads on individual vehicles can be approached by relaxing the evenly spaced headways pattern (rearrangement of departure times).
. resulting also in the reduction of wait and travel times. The load input for each vehicle is an average load profile that provides the measures of both passenger-km and empty-seat (space) km. This work presents the creation of public transport timetables with even average passenger loads on individual vehicles. cultural. This dynamic behavior can be detected through passenger load counts. educational. the passenger perspective.C. the maximum load at the critical point.) already said that: “Time is the most valuable thing one can spend”. and information provided by road supervisors. will eventually reduce one of the major sources of unreliable service. undoubtedly.
N. Rousseau. Daduna. Berlin. Springer-Verlag.M. Lecture Notes in Economics and Mathematical Systems 410... (Ed. I.I.H. (Ed. Computer-Aided Scheduling of Public Transport.. Daduna. J.M. Setting Frequencies on Bus Routes: Theory and Practice. N. J. Computer Scheduling of Public Transport 2. Berlin.. In Volmuller.N. H. (Ed. (Ed. A.H. In Wilson. Amsterdam.M. UNU Science Press.) (1981).) (1992). In Rousseau.References
Ceder. J.R. Proceedings of the 10th International Symposium on Transportation and Traffic Theory. Transportation Research 21A (1). North-Holland.M.) (1995). pp 1-7. P. O. Springer-Verlag. Berlin.) (1999). Wren.H. A. Transportation Research Board 818.R. Furth. NorthHolland. J. Computer-Aided Transit Scheduling. N. Wren (Eds. Branco. Springer-Verlag. pp 331-355. NorthHolland. Amsterdam. Rousseau (Eds.. Stern (1984). Desrochers.) (1985).M. Hammerslag (Eds. Koutsopoulos. Wilson. Wilson (1981).).): Computer Scheduling of Public Transport 2.. Bus Frequency Determination Using Passenger Count Data. Odoni. N. (Ed.) (1988). M. Berlin. A.P. J. pp 245-258.M. A. Ceder. Timetable Synchronization for Buses. Determination of Headways as Function of Time Varying Characteristics on a Transit Network. Methods for Creating Bus Timetables.): Computer-Aided Scheduling of Public Transport. Amsterdam.M. Springer-Verlag. Computer-Aided Transit Scheduling. A. J.H. Holland. pp 391-414.R.G. Urban
. Tal (1999). Ceder. Ceder. J. H.M. Lecture Notes in Economics and Mathematical Systems 386. Berlin. (1986). Wilson (1985). Paixao (Eds. Computer Scheduling of Public Transport 2: Passenger Vehicle and Crew Scheduling. Lecture Notes in Economics and Mathematical Systems 471. pp 59-83.. Optimal Transit Timetables for a Fixed Vehicle Fleet. A. Computer-Aided Transit Scheduling. (1984). Transportation Research 18A (5/6). pp 439-453. Lecture Notes in Economics and Mathematical Systems 308. A. Springer-Verlag.