An Application of Genetic Algorithm in Optimizing Jeepney Operations Along Taft Avenue, Manila

8th IEEE International Conference Humanoid, Nanotechnology, Nanotechnology, Information Technology Communication and Control, Environment and Management (HNICEM) The Institute of Electrical and Electronics Engineers Inc. (IEEE) – Philippine Section 9-12 December 2015 Water Front Hotel, Cebu, Philippines

An Application of Genetic Algorithm in Optimizing Jeepney Operations along Taft Avenue, Manila Raymund Paolo B. Abad

Dr. Alexis M. Fillone

Civil Engineering Dept. De La Salle University, Manila City, Philippines

Civil Engineering Dept. De La Salle University, Manila City, Philippines [email protected]

[email protected]

Dr. Elmer P. Dadios Manufacturing Engineering and Management Dept. De La Salle University, Manila City, Philippines [email protected]

Krister Ian Z. Roquel Civil Engineering Dept. De La Salle University, Manila City, Philippines [email protected]

 Abstract   — Effective transport service scheduling involves minimization of transit operating cost and maximization of generated revenue while meeting the demand for such service. This paper investigates the application of Genetic Algorithm in the optimization of Jeepney services along a busy section of their transit service routes. The main objective of this study is to improve transit service operations by minimizing travel and vehicle operating costs. The paper presented three types of services: Normal, Express A, and Express B. It was determined that a combination of the operation types would yield the smallest operating cost. Preference is still shown to Normal operation due to high passenger demand in the selected section during morning peak hours. Express services are more preferred during off-peak hours due to lower passenger demand.  Index Terms — genetic algorithm; transit scheduling; jeepneys

I. I NTRODUCTION Public utility jeepneys (PUJ) have become one of the most utilized mode of transport in Metro Manila, Philippines. In 2014 alone, about 36% of the generated trips inside Metro Manila have been made using jeepneys. The PUJ comes second to buses in terms of average occupancy (8.84  passengers) [1]. The success and popularity of jeepneys are attributed to local availability, intermediate size or capacity, accessibility, familiarity, economical fare, and extensive service networks [2][3][4][5]. The advantages of the PUJ makes it a highly viable alternative mode of a transport that even car-owners agree to use jeepneys in some cases [5]. In recent years, Metro Manila’s population is rapidly growing because of its development and urbanization. The growth of the metropolis put extensive pressure in the existing transport systems. In response to the growing demand of public transport services, the government has been taking steps to increase the capacity of mass transit in the form of the MRT, LRT, and PNR. Likewise, improved road transport services are also sought to meet the high trip demand. The jeepney sector is also pressured to adapt to the changing preferences of transit users for, although ‘Jeepneys have its merits, some commuters still consider them as “noisy”, “dirty”, and “dangerous”. They also associate negative images to jeepneys like “poor”, “stressed”, and “disorder” [5]. The negative perception of  jeepneys are brought by their their infamous road road behavior. behavior. PUJs are

978-1-5090-0360-0/15/$3 978-1-5090 -0360-0/15/$31.00 1.00 ©2015 IEEE

Fig. 1. Diagram showing undisciplined undisciplined Jeepney operations and and its impact on vehicular traffic

known to stop at the middle of roads to pick-up and drop  passengers which create instant bottlenecks at every crossing as shown in the diagram in Fig. 1 [6]. Likewise, the contribution of the PUJ to the worsening urban traffic situation is highlighted by their undisciplined undisciplined operation [7]. The changes in the transportation sector in Metro Manila  puts the role of jeepneys jeepneys in a transition transition role. From a main main mode of transport used in everyday travel, jeepneys are transformed into feeder modes. Despite this transition, it still holds a competitive advantage against other modes of transport considering its ability to access narrow roads or areas with limited transport service at inexpensive costs. However, this should not be a reason for jeepney services to remain incompetent. Jeepney operations are not subsidized by the government and its operations are highly vulnerable to the fluctuations of oil prices. Minimizing costs incurred by its operations should be the top priority of drivers and operators, alike. This paper addresses the minimization of costs incurred  by jeepney operations by applying Genetic Algorithm to optimize its service. Operations of jeepneys along Taft Avenue, Manila was chosen since it is a major thoroughfare in Metro Manila connecting three different cities: Manila, Pasay, and Parañaque. The study area will focus on the three kilometer section from Vito Cruz (Pablo Ocampo St.) to Ayala Boulevard intersection.

8th IEEE International Conference Humanoid, Nanotechnology, Information Technology Communication and Control, Environment and Management (HNICEM) The Institute of Electrical and Electronics Engineers Inc. (IEEE) – Philippine Section 9-12 December 2015 Water Front Hotel, Cebu, Philippines II. OPTIMIZATION PROBLEM

Fig. 2. The study area and common points of passenger exchange

This section was selected because it is a very busy corridor with frequent heavy traffic mainly due to the numerous trip generators in the area such as, higher educational institutions, schools, hospitals, government offices, and recreational areas (parks and malls), among others. There are many different jeepney routes that pass by along Taft Avenue indicating that there is a very high passenger demand in the study area. Table 1 shows the different operating routes and the number of jeepney units that pass by Taft Avenue. The table clearly shows that there is a need to effectively manage jeepney services along the chosen corridor  because of the extremely high number of jeepneys available to the public despite the short trip length.

Timetable development or dispatch scheduling of road based transport is not commonly done in Metro Manila. Most operators send out their whole fleet every day thinking it would maximize their generated income. This type of operation may make their revenues smaller because of greater operating costs. Sending out the entire fleet without knowing actual transit demand also has an effect to urban traffic flow since jeepneys would stubbornly wait for passengers along road segments. Efficient public transport operations must minimize waiting time and travel time of passengers and the costs for operators. Different operating schemes adapting to real-time traffic scenarios were formulated to address the optimization problem. In this paper, three types of operations were chosen: Normal, Express A, and Express B as shown in Fig. 3. In Fig. 3, red  points indicate that jeepneys should stop at that point and green  points indicate a skipped point. Normal operations allow the vehicle to run and stop along all common pick-up and drop-off  points. Express A and Express B operations allow the vehicles to stop at predetermined common pick-up and drop off points. Express A and Express B are different in terms of pick-up and drop-off points. Common pick-up points are used in this paper considering that jeepney operations do not have designated  pick-up and drop-off points. This paper would not employ zone operations considering that the selected study area is already a zone from the whole transit service route.

Fig. 3. Different operating types used in the study

TABLE I. JEEPNEY ROUTES OPERATING ALONG TAFT AVENUE Route MCU-Vito Cruz LRT Baclaran-Retiro (via L Guinto & Quiapo) Retiro-Libertad (via L Gunito & Sta Cruz) MCU-Pasay Rotonda (via Taft Ave) Escolta-Libertad (via L Guinto) P Campa-Libertad Kamuning-Vito Cruz LRT Monumento-Pasay Rotonda (via Taft) Libertad-Quezon Institute Blumentritt-Baclaran Blumentritt-Vito Cruz LRT Blumentritt-Remedios Project 6-Vito Cruz LRT Blumentritt-Pasay Rotonda Dapitan-Libertad (via L Guinto) Pasay Rotonda-Divisoria Cubao-Remedios Monumento-Vito Cruz (via Taft) Baclaran-P Campa (via L Guinto) Project 2/3-Remedios Baclaran-Dapitan (via L Guinto) Blumentritt-Baclaran Baclaran-Divisoria (via L Guinto) Monumento-Libertad (via Taft) Blumentritt-Libertad

Trip Length (km) 11.8 11.9 11.4 14.8 6.8 7.4 12.2 14.0 10.2 12.1 7.2 7.5 16.0 10.1 10.5 9.1 12.2 11.2 10.1 12.6 11.8 11.8 9.7 12.2 9.7

Total

III. METHODOLOGY

Units 12.0 4.0 28.0 71.0 3.0 1.0 22.0 28.0 19.0 261.0 23.0 10.0 10.0 56.0 31.0 23.0 323.0 11.0 67.0 48.0 182.0 185.0 247.0 35.0 27.0

1,727

The following sections show the employed methods in the model development and the algorithm formulation.  A.  Data Gathering Existing transit data were used in this study which were collected were done through frequency counts and on-board surveys. Frequency counts were done in order to determine which transport service routes heavily utilize the study area. The most jeepney routes that were counted operating along Taft Avenue were highlighted on Table 1. These routes contribute to about 60% of the jeepney volume along Taft Avenue [8]. Further analysis of the vehicle occupancies shows that jeepneys are, on the average, operate at 47% occupancy during the day. On-board passenger surveys were also conducted to determine the number of embarking and disembarking  passengers at common pick-up and drop-off points. This study optimizes the operation of jeepneys during the morning peak and off-peak periods. Fig. 4 shows the hourly average distribution of the boarding and alighting passengers along the

8th IEEE International Conference Humanoid, Nanotechnology, Information Technology Communication and Control, Environment and Management (HNICEM) The Institute of Electrical and Electronics Engineers Inc. (IEEE) – Philippine Section 9-12 December 2015 Water Front Hotel, Cebu, Philippines

Table 3 shows the description of the normal and express operations of jeepneys for the study area. Locations wherein  jeepneys would stop for passengers are designated with ‘1’, otherwise ‘0’.  B.  Assumptions Considering that jeepney operations are affected by a variety of factors, specific assumptions are to be laid out in this study. The assumptions include: (a) all jeepneys have the same capcacity; (b) jeepney operations are managed by a transit authority; (c) vehicles run at constant speed; (d) there are enough vehicles in the fleet to serve passengers; (e) vehicles would run even in the occurrence of traffic accidents and other unforeseen circumstances; (f) passenger arrival rate is uniform; and (f) the algorithm can be used in different frequencies or number of jeepney departures per hour. C.  Development of the Mathematical Model

Fig. 4. Passenger boarding and alighting during the peak period (top) and off peak period (bottom)

study area during the peak (6:00 AM – 9:00 AM) and off-peak  periods (9:00 AM – 12:00 NN). Table 2 shows the distances between points of embarkation and disembarkation along Taft Avenue and its corresponding travel times. An average travel velocity of 7.45 kilometers per hour (kph) was determined from on-board surveys during the study periods.

This study adapts the mathematical model from previous work wherein costs incurred due to passenger travel time and vehicle operating time is reduced [9]. The three objective functions consist of passenger waiting time, passenger travel time, and vehicle operating time as shown below. ⎛ r  j ⋅ hi2, j C 1 = VOT ∑∑ ⎜ ⎜ 2 i 1  j 1 ⎝   f 

 N 

=

=

S i

1, j ⋅

−

 ⎞ ⎟  ⎠  

hi , j ⎟

=

 f 

C 2

+

 N 

∑∑ [ L (t 

VOT 

i ,  j ⋅

(1)

) c] ( L

l  l   j + δ  i , j + δ  i , j −1 ⋅

+

i , j −

 Ai , j )⋅ δ  il , j ⋅ T 0

 

i =1  j =1

 f 

C 3

=

 N 

∑∑ [t  (δ  

VOC 

) c]

l  l  i , j + δ   i , j −1 ⋅

 j +

l  + δ  i , j ⋅

T 0  

(2) (3)

i =1  j =1

TABLE II. DISTANCE AND TRAVEL TIMES BETWEEN COMMON POINTS OF EMBARKATION AND DISEMBARKATION Location

Distance (km)

t (mins)

0.5 1 1.2 1.4 1.7 1.9 2.2 2.4 2.7 3.1

4.027 4.027 1.611 1.611 2.416 1.611 2.416 1.616 2.416 3.221

Vito Cruz - Estrada Estrada - Quirino Quirino - Remedios Remedios - Malvar PWU Malvar PWU - Pedro Gil Pedro Gil – PGH PGH - Faura Faura - NBI/UN  NBI/UN - Kalaw Kalaw - Taft Ayala

 j

TABLE III. OPERATION TYPE AND CORRESPONDING STOP ASSIGNMENTS Location Vito Cruz Estrada Quirino Remedios Malvar/PWU Pedro Gil PGH Faura  NBI/UN Kalaw Taft Ayala

Peak

Where, i – vehicle, i = 1,2,…,f;  j – passenger embarking and disembarking point along the route, j = 1, 2,…,N l   – form of operation, l   = 1 means normal operation, l = 2 means express A operation, l   = 3 means express B operation hi,j  – headway between vehicle i-1 at stop j and computed as:

Off Peak

 Normal

Express A

Express B

Express A

Express B

1 1 1 1 1 1 1 1 1 1 1

1 0 0 0 1 1 0 1 1 0 1

1 0 1 0 0 0 1 0 0 1 1

1 0 1 0 1 1 0 1 0 0 1

1 0 1 0 1 0 1 0 0 1 1

hi , j

=

h+

∑ (δ   (T  l  i , k 

k =1

0 +

 j

∑ (δ  

c )) −

(T 0 + c )) 

l  i −1, k 

(4)

k =1

r  j – arrival rate at stop j S i,j  – the total left passengers of vehicle i at stop  j  and computed as:  N 

S i , j

=

∑ s

i , jk 

  T 0 – dwell time and calculated as T0 = 7.0401797Emb + 1.76214Dis [10]  Emb – number of boarding passengers at stop j  Dis – number of alighting passengers at stop j k = j +1

(5) (6)

8th IEEE International Conference Humanoid, Nanotechnology, Information Technology Communication and Control, Environment and Management (HNICEM) The Institute of Electrical and Electronics Engineers Inc. (IEEE) – Philippine Section 9-12 December 2015 Water Front Hotel, Cebu, Philippines l 

δ  i,j  – “0-1” variable. 1 when the vehicle i  stops at  j ,

otherwise 0. l  δ  i  – operation type of vehicle i c – acceleration and deceleration time (0.0833 minutes) VOT – value of time of passengers (P78.1/hour) [11] VOC   – vehicle operating cost (calculated as P1.82/min) [11] The first objective function takes into account the waiting costs of passengers for a service vehicle to arrive. It includes the waiting time of the vehicle and the time caused by skipping stops. The second objective function calculates the onboard costs of passengers. Finally, the vehicle operation cost is calculated from the travel time, acceleration and deceleration time, and dwelling time. Dwell time per stop were determined from the regression equations developed from [10]. Meanwhile, VOT and VOC are calculated from the study reports of [11].

 A.  Initial Population The initial population is randomly generated. The length of coding is defined by the minimum and maximum headway. The chromosome is composed of n number of genes where n is the frequency of jeepneys within an operating period (one hour for this study). A gene is represented by the type of operation in the study area. Normal type of operations is represented by 1. Express A and Express B operations are represented by 2 and 3, respectively. The number of genes in each chromosome is held constant for each determined headway. This study tested four different headways resulting to different operating combinations.  B.  Fitness Function This study minimizes the total costs incurred during the operation. The solution provided should have the least waiting and travel time for the passenger. The solution that satisfies the equation below will be considered the fittest individual: min X = f(X) = C 1 + C2 + C3  (7)

IV. GENETIC ALGORITHM In this paper, genetic algorithm was used to find a solution to the optimization problem. Genetic algorithm is an adaptive search technique patterned after the evolutionary process of selection and “survival of the fittest” [9]. Genetic algorithm is very powerful in finding solutions from complex problems. It  begins with an initial group of strings (population) that represents a possible solution to the problem. A chromosome, represented by a string or a member of the population, is evaluated by a fitness function. From the evaluation, chromosomes are deemed whether fit or not to be mated. Each chromosome would generate a new population of chromosomes. The generation of a new population is often characterized by three processes: reproduction, crossover, and mutation. Generation of new populations is continuous until convergence or a local minimum or maximum is attained. The search algorithm is terminated once convergence is attained. The process of the genetic algorithm in this study is shown in Fig. 5.

C. Genetic Operators 1) Selection of Parents Roulette gambling law was chosen as the selection method for individuals. Roulette gambling law selects chromosomes in random. Chromosomes that are more fit have a higher  probability of being selected. 2) Crossover Crossover is the operation that generates a new set of  population [12]. Crossover works by crossing two individuals at a certain location. One-point crossover was used in this study as shown in Fig. 6.

Fig. 6. Example of one-point crossover between two parents

3)  Mutation Mutation works by altering a specific gene from a selected chromosome. Chromosomes are mutated at random locations when it meets a defined mutation probability.  D.  Definition of Parameters

Fig. 5. Flowchart for the Genetic Algorithm

Other parameters that are needed in the model are described in this section. Acceleration and deceleration time of jeepneys are estimated as 0.0833 minutes (5 s) considering the short distances between common points of passenger exchange. Mutation rate is taken as 0.005, population size and number of generations is kept constant at 20 and 100, respectively. Four headways of 3, 4, 6, and 10 minutes for an hourly operating  period were tested. Other parameters can be found in Tables 2 3.

8th IEEE International Conference Humanoid, Nanotechnology, Information Technology Communication and Control, Environment and Management (HNICEM) The Institute of Electrical and Electronics Engineers Inc. (IEEE) – Philippine Section 9-12 December 2015 Water Front Hotel, Cebu, Philippines

V. R ESULTS AND A NALYSIS The main objective of the algorithm is to create an operating schedule with the minimum cost. Hence, this algorithm should minimize the value of the fitness level for each passing generation. Fig. 7 shows that convergence is attained after a number of generations during the peak period with a frequency of 20 jeepneys for the study period.

skip different stops in order to minimize total travel time. It also suggests consecutive vehicles having the same operating type to minimize headways. Passengers waiting at stops are also smaller which explains the preference of express services during off-peak hours. However, on larger headways,  preference is again given to normal operations. TABLE V. OPTIMIZED OPERATING COMBINATIONS Frequency

Operating Combination

Fitness

 Morning Peak Period

6 10 15 20

311112 3111133232 311111222222222 21111111133222232322

6 10 15 20

21112 2311111133 333333323333333 33333323333333323332

2563.613 2025.147 1971.975 2182.294

 Morning Off Peak Period

Fig. 7. Fitness for every generation (f=20, peak period)

Convergence was also attained for other conditions and variables as shown in Table 4. It can be observed that as the length of the chromosomes grow longer, the higher number of generations were needed to attain the minimum value. Convergence to a minimum value shows that genetic algorithm finds the desired fitness. TABLE IV. CONVERGENCE OF FITNESS VALUES ON DIFFERENT PARAMETERS Frequency 6 10 15 20

AM Peak 10 20 85 60

AM Off-Peak 10 35 80 95

The combination of operating types are shown in Table 5. Each vehicle is given a headway depending on the number of assigned vehicles within the operating period (1 hour). The combination of operations would vary depending on the  passenger demand for each location along the route. The algorithm has calculated a viable operating combination at the most economical cost. It is seen in Table 5 that during the peak periods, preference is given with consecutive similar operating types. This occurrence is due to the fact that the model depends on the least operating time between points of embarkation and disembarkation. Headways are normally smaller when consecutive vehicles have the same operating type. Furthermore, Express services increase the amount of waiting  passengers which increase passenger waiting costs. Considering that the selected study area has high passenger volume at every stop, the program tends to prefer the normal type of operations. Off-peak operating combinations show a different trend especially at smaller headways. It shows that given the smaller passenger demand, jeepneys are suggested to

2702.958 2238.627 2171.11 2324.204

VI. CONCLUSION This paper showed that genetic algorithms can be used to optimize the operation of public utility jeepneys (PUJs) along a  busy corridor of their service route. A model was adopted so that waiting and travel time of passengers and operating time of  jeepneys are minimized. The use of genetic algorithm in the minimization problem was proven feasible by showing its convergence at a local minimum after a number of generations. The developed model highly depend on the headways  between vehicles and the number of passengers boarding and alighting the vehicle at designated stops. At sections wherein  passenger demand is relatively high, preference was given to normal operations to address the passenger waiting costs. Express services are highly preferred during off-peak hours. The inclination towards express services is a result of the low  passenger demand at these times. It was also shown in both study periods that consecutive vehicles having the same operating type further minimizes the costs by maintaining the headways. The variations in passenger demand and headways should be monitored as these factors serve as the reference of the operating combinations that will be implemented in the operating period. Overall, the application of genetic algorithm in optimizing  jeepney operations along a busy section of a transit service route is possible. It should be the responsibility of the respective transit agencies to select which stops or locations would be served or skipped. Incorrectly determined skipped stops would normally make express services more expensive and less desirable in the algorithm. This paper proves the high capability of genetic algorithms in finding the optimal operations of transit services considering different costs. ACKNOWLEDGMENT The authors express their gratitude to the support given by Dr. Alexis Fillone and the PUBFix project team which is supported by the Philippine Council for Industry, Energy, and

8th IEEE International Conference Humanoid, Nanotechnology, Information Technology Communication and Control, Environment and Management (HNICEM) The Institute of Electrical and Electronics Engineers Inc. (IEEE) – Philippine Section 9-12 December 2015 Water Front Hotel, Cebu, Philippines

Emerging Technology Research and Development of the Department of Science and Technology (DOST-PCIEERD).

[6] Gitano-Briggs, H. “Future Transport: A Review. Motivating Factors. Technological Trends. Analysis of RP Prospects. (Jeeps Tricycles)

R EFERENCES

[7] Chiu, C. L. “Lessons from Jeepney Industry in the Philippines” International Review of Business Research Papers, Vol. 4, No. 1, pp. 68-84, 2008

[1] Department of Transportations and Communications (2015, July 28) “The Project for Capacity Development on Transportation Planning and the Database Management in the Republic of the Philippines (MUCEP) P roject Update 2” [2] Bayan, J. M. “Cost Characteristics of Bus and Jeepney Transport Systems in Metro Manila”, University of the Philippines, College of Engineering, Quezon City [3] Ebata, J. “Jeepney Business in Metro Manila: What are the conditions for its sustainability?”, University of the Philippines, College of Engineering, Quezon City [4] Labastilla, G.L., and Viloria, O., “Empirical analysis of Route Operating Characteristics, Level of Service and Profitability: The Case of Jeepneys in the City of Manila”, Proceedings of the Eastern Asia Society for Transportation Studies, Vol. 2, pp. 201217, 1999 [5] Okamura, T., Kaneko, Y., Nakamura, F., and Wang, R., “Passengers’ Attitudes to the Service items of Jeepneys in Metro Manila by Different Lifestyles”, Proceedings of the Eastern Asia Society for Transportation Studies, Vol. 9, 2013

[8] Department of Transportation and Communications (2012) “Road Transit Rationalisation Survey” [9] Sun, C., Zhou, W., and Wang, Y. “Scheduling Combination and Headway Optimization of Bus Rapid Transit”, Journal of Transportation Systems Engineering and Information Technology, Volume 8, Issue 5, October 2008 [10] Roquel, K.Z.R., and Fillone, A. “Current Public Transport Service Operating Characteristics along Epifanio delos Santos Avenue (EDSA), Metro Manila”, Proceedings of the Eastern Asia Society for Transportation Studies, 2015 [11] Japan International Cooperation Agency (2014) “Roadmap for Transport Infrastructure Development for Metro Manila and its Surrounding Areas (Region III and Region IV-A” [12] Dadios, E.P., and E.A.B., F. (1995) “Genetic Algorithm for Job Scheduling Problem using Three-Parent Crossover with Invalid Chromosomes Repair Strategies. IEEE Computational Intelligence.