GIS-Based Optimization for Gas Distribution Route Design: A Case Study of PT Gagas Energi Indonesia

PT Gagas Energi Indonesia is a company engaged in the trading and natural gas business, one of which is Gaslink C-Cyl. Route determination for Gaslink C-Cyl distribution is carried out subjectively based on the proximity of locations between customers, thus creating inefficient routes, affecting the number of vehicles needed, and causing high transportation costs. Based on this problem, this study aims to minimize transportation costs by determining the optimal route and number of vehicles needed using the Multi-Trip Vehicle Routing Problem Time Windows Simultaneous Pickup and Delivery (MTVRPTW-SPD). This problem is solved using Geographic Information System (GIS) based optimization. To evaluate GIS-based optimization performance, comparison between existing condition and optimization are done for 50 demand points. Then, two scenario was developed, 100 and 200 demand points to explain the decision implication related to demand uncertainty. Based on the optimization process for 50 demand points gives better solution than existing condition in term of number of fleets needed, total distance travelled, and total transportation cost which result 3 units, 46,5%, and 43,5% respectively.


Introduction
Distribution process is a complex problem because it involves several parties, source (depot) and several destinations.The more nodes-both depot and destinations will affect the performance of distribution.It can be total distance travelled, distribution cost, or etc. PT Gagas Energi Indonesia is a subsidiary of Pertamina Gas Negara (PGN), which is engaged in trading and natural gas, one of which is Gaslink C-Cyl.Gaslink C-Cyl, which contains CNG gas, is distributed to customers; empty Gaslink C-Cyl cylinders will first be filled with CNG gas at the nearest Gas Refueling Station (SPBG) from the customer's area.
Figure 1 shows that one of the SPBGs with the highest utilization each year is the Klender SPBG (Depot) which is one of the gas suppliers for the DKI Jakarta area.Based on this, the Klender SPBG is a case study of this research because it has the most potential number of potential Gaslink C-Cyl customers, as seen from the high utilization of SPBG from year to year compared to other SPBGs.

Figure 1. Gas balance and SPBG utilization of PT Gagas Energi Indonesia
In this case, finding the optimum number of vehicles and route are crucial because it will affect the total transportation costs.The transportation costs are affected by number of vehicles used, personnel costs-driver and assistant, and fuel consumption.While in the existing condition, the routes are determined subjectively by considering distance between each customer.This solution does not guarantee the optimality.It causes the number of vehicles rented and the total distance are high.
During the transportation, the delivery and pickup for empty tube are done simultaneously.The vehicles can take several routes as long as they do not violate time windows.Besides, it also considers some restrictions, such as the customers demand must be met, the vehicle capacity, the start and end trip must be at depot.This problem classified as MTVRPTW-SPD.The MTVRPTW-SPD model is combination of MTVRPTW and VRP-SPD.MTVRPTV is extended from VRPTW by allowing multiple trips for vehicles during time horizon (Huang, Li, Zhu, & Qin, 2021) and (Cattaruzza, Absi, & Feillet, 2016).While VRP-SPD is the extension of VRP where the customers have both a pickup and a delivery demand (Koç, Laporte, & Tükenmez, 2020) and (Redi, et al., 2021).In this paper, the term trip is used to describe the occasion where a given vehicles leaves the depot to visit customers and return for several times without exceeding the time windows.While the term route refers to the sequence of customer visit indicated by a path that a vehicle follows during a trip (Cueto, Gjeroska, Vilalta, & Anjos, 2021).
Based on this problem, this paper aims to determine the optimum distribution routes and number of vehicles used to minimize total transportation cost.To solve this problem, a GIS-based optimization is implemented by adopting MTVRPTW-SPD model.A network analyst tool in GIS is used to design route for CNG distribution.The problem used is based on the real case where the problem is large and based on spatial information.
The gas distribution through cylinder research has been widely studied before.The main characteristic of this problem is pickup and delivery carried out simultaneously and it has been studied by (Singamsetty & Thenepalle, 2021) and (Panicker & Mohammed, 2018).Garside & Laili (2019) implemented Cluster First Route Second (CFRS) heuristics to solve Multi-Trip Periodic Vehicle Routing Problem (MTPVRP).They solved a real case study of Liquefied Petroleum Gas (LPG) distribution to minimize total cost, which is the sum of total vehicle fixed cost and total travelling cost.Yuliza, Puspita, Yahdin, & Emiliya (2020) deal with LPG distribution using hybrid Clarke and Wright Algorithm and LINGO to determine optimum route that provide minimize total distance by implementing Capacitated Vehicle Routing Problem (CVRP).
Many previous research utilized GIS-based optimization method to solve optimization problems include VRP and its variants.ÖZCEYLAN, KOÇ, & ERBAŞ (2018) solve Multi-Trip Heterogeneous Fleet Vehicle Routing Problem (MTHFFVRP) using a GIS-based optimization, based on a tabu search algorithm with a real case of retail chain at Turkey.The results show that GIS-based optimization obtain good results in term of number of fleets, vehicle capacity utilization rate, total distance, total travel time, total en-route time, and computation time for a rich VRP characteristic.(Hashi, Hasan, & Zaman (2016) figure out the school bus routing and scheduling to optimize the overall transportation cost by minimizing the number of fleets and time spent.They 0 1.000.0002.000.0003.000.0004.000.0005.000.0006.000.0007.000.0008.000.000 Gas Balance and SPBG Utilization of PT Gagas Energi Indonesia implement GIS based solution to solve this problem.The GIS based solution provides a better solution compared to the existing semimanual system.GIS is getting popular to address strategic application where data (spatial and nonspatial) are used as an input.In addition, GIS has more advantages in the collection, analysist, and visualize the data (Krichen, Faiz, Tlili, & Tej, 2014).This makes the data presented more attractive and helps the user to interprets data patterns faster than plain text (Rosita, Falahah, & Sanjaya, 2017).ArcGIS 10.8 software uses Djikstra's shortest algorithm to find the shortest path between two points to create a distance matrix so that it is possible to find solutions faster than other algorithms (Rachmawati & Gustin, 2020).In addition, another algorithm used in ArcGIS 10.8 for Optimization is the tabu search algorithm, which is the most widely used metaheuristic method (Faiz, Krichen, & Inoubli, 2014).This method can work by preventing repetition in each iteration, allowing users to obtain local optimal solutions using tabu lists and neighbor generation in reasonable computing time (Faiz, Krichen, & Inoubli, 2014).The rest of this paper describes about research methods conducted.It includes mathematical model that represents the problem, the data used and procedure to collect the data, and the steps to do this research.Then, we explain and analyze the results in the result and discussion section.The last, draw the conclusion and give the recommendation for future research are mentioned in the conclusion section.

Methodology
The system under description is a distribution system of LNG cylinder tube that deliver and pick up the product from customers simultaneously.The delivery and pickup are proceeded during the time windows.The vehicles used homogenic.When the delivery and pickup demand exceed the vehicle capacity, the vehicle is allowed to have multi trips as long as within its time windows.In this case, the vehicles will back to depot first to unload the empty cylinder and load the full cylinder to meet the rest of customers' demand.The mathematical model used refers from (Kenaka & Suprayogi, 2021).Here is the complete mathematical model of this problem:  The objective function is to minimize total transportation cost-fixed cost and variable cost defined in Eq. (1).Constraints (2)-(3) describe that each vehicle must start and end at depot.Constraint (4) guarantees the route continuity.Constraint (5) ensures the service time at each demand point.Constraints ( 6)-( 11) make sure the time windows for demand point and depot are not violated.Constraint (12) restricts the loading time at depot.The route sequence is defined by constraints ( 13)-( 17).Constraint (18) describes the delivery load when a vehicle returns at the depot.Flow conservation for delivery quantity at each customer is stated at constraint (19).The total load from depot and the vehicle capacity constraint are defined at constraints ( 20) and ( 21) respectively, constraints ( 22)-( 27) define the characteristic of decision variables.
In this research, we compare the existing condition and the result of GIS-based optimization for 50 demand points then we develop two scenarios for the larger demand point-100 and 200 demand points to illustrate the uncertainty of the demand itself.The determination of number of demand point is based on the minimum, average, and maximum demand point from expert judgement.
This research is solved following six steps.First, define a network dataset.Second, run vehicle routing problem in Network Analyst extension.The several examples of Network Analyst Input Parameter Interface can be seen at Figure 2(a)-(c).To solve the problem, we used the primary and secondary data.The data collected are maps of Jakarta, road network, set of location (depot and customers), attribute of location, attribute of vehicle, cost component, and processing time.The vehicle attribute, cost component, set of locations, and demand data are shown at Table 1, Table 2, and Table 3 respectively.Third step is verification and validation, verification is carried out by identifying the error in GIS, while validation is done by checking the result that it is not violated the constraints.The validation is made by comparing the GIS results and manual computation related to total time, total cost, and vehicle capacity constraint in Table 4-6.Fourth, we compare the existing condition and GISbased optimization results to see the efficiency of the model proposed.Fifth, develop two scenarios to see the implication of the demand uncertainty so the managerial implication can be carried out based on the real situation.Finally, we draw a conclusion under limitations of our research and give potential future research.

Results and Discussion
In this section we compare the result based on GIS and existing condition which can be seen at Table 7 and Table 8 respectively.The routes representation of GIS-based Optimization is shown in Figure 3.
Based on Table 7 and Table 8 we can see that the optimization results give the better solution in terms of number of vehicles used, total distance, and total cost.The existing condition for delivery and pick up 50 demand points need seven vehicles while in the optimization only need 4 vehicles.Because in the optimization model, we allowed multi-trips.So, the vehicles are allowed to return to the depot and start deliver again if the time windows are not violated.This policy benefits the company to minimize the vehicle rent cost and it will impact the total transportation cost.The total cost decreases about 2.605.798rupiahs.This condition is affected by some factors, such as the number of vehicles used and the total distance.The number of vehicles used has a linear relation to the total cost.The least number of vehicles used, the smaller total cost incurred, as well as the total distance travelled.
Finally, we compare the result from different number of demand points.The greater number of demand points lead to higher number of vehicles need to deliver and pick up LNG.The number of demand points also affect the total distance travelled.they have a linear relation.Based on this optimization, the managerial implication for the company is determining the optimum routes is one of the key factors to minimize the total transportation cost.Based on the expert judgement, the demand points are between 50 to 200, in this case we can suggest the company to rent between 5-11 trucks depend on the number of demand points.Besides, the other important factor is the transportation policy.In this result, we can conclude that by implementing multi-trips is more beneficial for the company than doing single trip.By allowing the vehicles to have multi trips, it will rise the time utilization for the vehicle.The company should consider the safety aspect for the worker (drivers and their assistant).However, in the real case the company already mitigate this condition by assigning two drivers and an assistant in each vehicle.

Conclusion
In this research, a GIS-based Optimization is used by adopting MTVRPTW-SPD model.Implementing this model is beneficial for the company because it allowed the vehicles have several trips in a day.The model proposed can reduce the number vehicle used and the total distance.This condition leads to improve the total transportation spent.
Based on the optimization process for 50 demand points, it reduces the number of vehicles, the total distance, and the total cost about 3 unit, 46,5%, and 43,5% respectively.Other finding is if the demand points at the range 50-200 the company should consider vehicle rented about 4-11 trucks, with the detail at the Table 10.
The limitations of this research are considering the average of vehicle speed and not considering the road restriction.For the future research, the vehicle speed can be used as one factor to generate the scenario and developing model that considering road restriction.

Table 1 .
Vehicle specifications

Table 7 .
Existing Condition for 50 Demand Points N o

Table 8 .
GIS-based Optimization for 50 Demand Points

Table 10 .
Recapitulation for Each Scenario