AUCTION BASED DISPATCH MODEL IN ONLINE MOTORCYCLE TAXI SYSTEM

Transcription

1 AUCTION BASED DISPATCH MODEL IN ONLINE MOTORCYCLE TAXI SYSTEM PURBA DARU KUSUMA School of Electrical Engineering, Telkom Univerity, Indoneia ABSTRACT One problem in the exiting online motorcycle taxi ytem i generalization among driver. The generalization include price, travel ditance, and pickup ditance. Meanwhile, both driver and paenger cannot be generalized. For example, ome driver want to get higher price while other driver want to take horter travel ditance. In the other ide, ome paenger want to get lower price while other paenger want to get lower waiting time. Baed on thee need, in thi reearch, we propoe new online motorcycle taxi dipatch ytem that accommodate thi variou requirement. The propoed model i developed baed on auction model. The auction i done automatically, ealed, and it i ingle round auction. In thi reearch, the driver requirement are: maximum travel ditance, maximum pickup ditance, and price range. Meanwhile, the paenger requirement are: maximum waiting time and price range. There are three propoed model in thi reearch. In the firt model, pickup ditance limitation i implemented. In the econd model, travel ditance limitation i implemented. In the third model, both pickup ditance and travel ditance limitation are implemented. In the tet, beide comparing to each other, thee propoed model are alo compared with the previou nearet driver model. The tet reult i a follow. The previou nearet driver model perform the highet ucce ratio. The firt model perform the highet average driver revenue. The third model perform the lowet average waiting time and average pickup ditance. Keyword: Dipatch ytem, Online motorcycle taxi, Auction, Nearet driver, Multi agent. 1. INTRODUCTION Nowaday, online motorcycle taxi buine grow very fat, epecially in Indoneia. At the beginning, there are many online motorcycle taxi provider. Now, the competition i jut two hore race between Go-Jek with it Go-Ride and Grab with it Grab-Bike [1,2]. Go-Jek i local company while Grab i Malayian baed company. Before thi, Uber i in the competition. After the very hard competition, Uber decided to ell it buine in outh-eat Aia to Grab [1]. Thi action i caued by the fierce competition in ride hailing indutry, epecially in Indoneia [3,4]. The buine that i old to Grab include the raid hailing and food delivery [4]. Thi fierce competition that i faced by Uber make the company lot at $4.5 billion in 2017 [4]. So, by conolidating buine, it hope the company will make profit a oon a poible [4]. Even the growth i very fat and the public acceptance i very high, the buine model in online taxi i till far from table. The ytem, uch a the dipatch ytem, reward ytem, and compenation ytem are till improved continuouly. But, there i big difference between thee two provider. Go-Jek adopt multi agent dipatch model which the driver can accept or reject the order [5]. Thi agent baed model ha been imulated in our previou work to meaure it performance [5]. Meanwhile, Grab adopt mandatory dipatch model which the driver mut accept all order. Thi mandatory model i alo ha been imulated in our previou model that ue nearet driver model to meaure it performance [6]. The problem in the both exiting model i that thee model tend to generalize both the driver and the paenger [5,6]. Epecially in the nearet driver method, the order will be allocated to the nearet available driver [6]. The purpoe i by allocating the order to the nearet driver, paenger waiting time and driver pickup ditance will be the lowet [7]. It i becaue in many reearche, one of taxi paenger interet i in reducing waiting time [7]. Meanwhile, one of taxi 6134

2 driver interet i reducing pickup ditance becaue in taxi buine model, pickup cot i not charged to the paenger and thi cot reduce driver revenue [8]. Unfortunately, both paenger and driver interet cannot be implified a it i covered in the exiting online motorcycle dipatch model [5,6]. After large number of converation with the driver and paenger, ome intention have not been covered in the exiting ytem. Some driver want to take horter travel ditance order while other are tolerant to take longer one. Some driver want to take horter pickup ditance order while other are tolerant to take longer one. Some driver want to take higher travel price while other are tolerant to lower one. Thi condition alo occur in paenger ide. Some paenger want to get lower waiting time while other are tolerant to longer one. Some paenger want to get lower travel price while other are tolerant to higher one. That i why ome paenger, for empathy reaon, give ome tip to driver. Baed on thi condition, the motivation and the jutification of thi reearch are propoing and to developing new dipatch model that accommodate thoe mentioned interet. Thi model alo treat both driver and paenger a a peronal and not jut a a common entity [6]. Auction baed approach i propoed in thi reearch becaue many exiting reearche in taxi dipatch model ued nearet driver approach [7,14,15], FIFO approach [16,17], or agent baed approach [5]. Baed on the exiting condition and the reearch motivation, the main reearch quetion i how thi new model i developed or in what bai the model will be developed. The following quetion i how better the propoed model olve thi problem comparing to the exiting dipatch model. So, the reearch purpoe i to develop the new online motorcycle taxi dipatch model that accommodate thee driver and paenger need. The propoed model then will be compared with the exiting model, epecially nearet driver model [6]. The comparion i needed to meaure how better the propoed model compared with the exiting model. The hypothei of thi reearch i thi auction baed dipatch model i better than the previou nearet driver dipatch model. To prove thi hypothei, thi propoed auction baed model then will be compared to the nearet driver baed model both for driver and paenger interet. The parameter that are evaluated include: ucce ratio, revenue, pickup ditance, and waiting time. In thi reearch, the propoed model i developed by combining auction ytem with multi agent ytem. The auction ytem i ued becaue a part of negotiation ytem, the purpoe of the ytem i finding bet olution among partie. Meanwhile, multi agent ytem i ued becaue thi ytem i broadly ued in model that treat entitie inide the ytem a autonomou object. Baed on the reearch purpoe and method that i ued for the bai, contribution of thi reearch are a follow. Firt, thi reearch propoe peronalization approach rather generalization approach a it i ued in many taxi dipatch model. Second, thi reearch alo include price range in dipatch proce rather than common parameter, uch a: pickup ditance, waiting time, and idle time. Third, thi reearch enriche the implementation of computational baed auction, epecially for taxi dipatch proce. Baically, thi reearch poition i the combination between IT reearch and computing reearch, but, the IT portion i more dominant rather than the computing reearch. In IT area, thi reearch i the part of the online motorcycle taxi ytem o that the ytem integrate the three entitie: paenger, driver, and online motorcycle taxi company. In the entire online motorcycle taxi ytem, thi model can be ued in the dipatch part. Meanwhile, there i cro ection with the computing area becaue there i effort to develop auction model automatically. The paper i organized a follow. In the firt ection, we decribe the background, reearch motivation, reearch quetion, reearch purpoe, and the paper organization. In the econd ection, we explain the exiting online motorcycle taxi dipatch ytem. In the third ection, we explain the baic concept of the negotiation and auction. In the fourth ection, we explain the propoed model. In the fifth ection, we explain the implementation of the model into the imulation application. In the ixth ection, we explain the the tet, the reult analyze, and the reearch finding. In the eventh 6135

3 ection, we make concluion and propoe future reearch potential. 2. EXISTING ONLINE MOTORCYCLE SYSTEM In thi ection, we explain the condition in exiting online motorcycle taxi ytem. In thi ytem, we will how the mandatory baed ytem that i hown in our previou work [6]. There are two method: the nearet driver model and longet driver model. In the nearet driver model, the order will be dipatched into the driver whoe location i the nearet to the pickup location [6]. The purpoe of thi model i to minimize the paenger waiting time and the driver pickup ditance. The econd model i the longet idle time. In thi model, the order will be allocated to the driver within the obervation range whoe idle time i the longet one [6]. The purpoe of thi model i to prioritize the driver who ha waited order for the longet time. The obervation range i ued to keep the paenger waiting time and driver pickup ditance till low. The illutration i a follow. Suppoe that there are 10 driver around the paenger who make pickup order. The driver parameter are ditance between paenger and driver (d pa ) and driver idle time (t idle ). The obervation range i et 3 kilometer. The detail information i hown in Table 1. Table 1. Driver Current Parameter Driver d pa (km) t idle (econd) m m m m m m m m m m Baed on data in Table 1, when the nearet driver model i implemented, then the ytem will allocate the order to driver m 7. It i becaue the ditance between the paenger pickup location to the driver m 7 location i the nearet among other driver. The problem, the idle time of driver m 7 i the lowet one among other driver. So, from queuing point of view, the nearet driver model i not o fair. Meanwhile, when the dipatch ytem adopt longet idle time model [6], the cenario i a follow. When the obervation range i 1 kilometer, the order will be allocated to driver m 5. When the obervation range i 2 kilometer, the order i till allocated to driver m 5. When the obervation range i 3 kilometer then the order will be allocated to driver m 1. When the obervation range i 4 kilometer, the order will be allocated to driver m 2. When the obervation range i 5 kilometer, the order i till allocated to driver m 2. If it i aumed that the driver peed i 2 kilometer per minute, then the maximum paenger waiting time from obervation range 1 kilometer to 5 kilometer i 2 minute, 4 minute, 6 minute, 8 minute, and 10 minute conecutively. The problem i there are condition that are till uncovered by the exiting ytem [5,6]. The firt problem i the driver maximum pickup ditance (d maxpick ) and maximum travel ditance (d maxtrav ). Some driver tend to get order from horter pickup ditance or travel ditance. Other driver are tolerant for further pickup ditance or travel ditance. The example of driver maximum pickup ditance and maximum travel ditance parameter i hown in Table 2. Table 2. Driver Ditance Parameter Driver d maxpick (km) d maxtrav (km) m m m m m m m m m m Now, there are two paenger named p 1 and p 2 who make travel order. When he make an order by uing hi mart phone, hi pickup and detination location are ent to the ytem. Then, the pickup location relative to the driver and the travel ditance i calculated. The p 1 travel ditance i 11 kilometer and the p 2 travel ditance i 6 kilometer. The pickup ditance data of thee paenger relative to driver i hown in Table

4 Table 3. Driver Ditance Parameter Driver d pick (p 1 ) (km) d pick (p 2 ) (km) m m m m m m m m m m By comparing data in Table 2 and Table 3, the relation i a follow. For paenger p 1 pickup ditance parameter, the candidate that can be allocated are {m 2, m 3, m 8, m 9, m 10 }. For paenger p 2 pickup ditance parameter, the candidate that can be allocated are {m 1, m 2, m 4, m 5, m 6, m 8, m 9, m 10 }. Another parameter i the maximum travel ditance. Baed on maximum travel ditance, the driver that can get the p 1 order are {m 3, m 4, m 10 } while the driver that can get the p 2 order are {m 1, m 2, m 3, m 4, m 7, m 8, m 9, m 10 }. By comparing the pickup ditance and the travel ditance parameter, the driver that can accept the p 1 order i {m 10 } while the driver that can accept the p 2 order are {m 2, m 4, m 8, m 9, m 10 }. The next problem i the price. There are two type of price: the target point (p target ) and the reervation price (p re ). Thi price i for both the driver and the paenger. The next illutration i the example of the price problem. Table 4. Driver Price Parameter Driver p re (rupiah/km) p target (rupiah/km) m 1 1,500 5,000 m 2 3,000 4,000 m 3 2,500 4,500 m 4 1,000 2,000 m 5 1,500 3,000 m 6 3,500 5,500 m 7 2,500 3,000 m 8 2,000 3,200 m 9 2,500 3,500 m 10 2,000 2, NEGOTIATION AND AUCTION Negotiation i a proce to find olution between partie. Negotiation i needed to olve problem which deciion or olution cannot be determined by one party without accommodating other partie interet. In it baic form, there are three common term in negotiation proce: target point, reervation price, and zone of poible agreement (ZOPA) [9,10]. Zone of Poible Agrrement i alo called a ettlement zone [11]. Target point i the ideal point for a party. Reervation price i the wort point where party till can make agreement [9]. Reervation price i alo called a reitance point [11]. ZOPA i the area or range that the agreement will be done in it. The illutration i hown in Figure 1. Suppoe that there are two paenger: p 3 and p 4. The p 3 target point i 1,000 rupiah per kilometer and the reervation price i 2,000 rupiah per kilometer. The p 4 target point i 1,500 rupiah per kilometer and hi reervation price i 3,000 rupiah per kilometer. Back to the et of the driver, the driver target point and reervation price are hown in Table 4. By comparing the price range between driver and paenger, the reult i a follow. For paenger p 3, the candidate that can get the order are {m 1, m 3, m 4, m 5, m 8, m 10 }. For paenger p 4, the candidate that can get the order are {m 1, m 2, m 3, m 4, m 5, m 7, m 8, m 9, m 10 }. The rationale i the candidate reervation price mut be equal to or le than the paenger reervation price. Figure 1. Negotiation Area Baic Concept Illutration in Figure 1 how the relation between negotiation term. Thi illutration i ingle party. Common example of negotiation i the price negotiation between merchant and buyer. The merchant goal i maximizing the price while the buyer interet i minimizing price. So, in thi example, the merchant target point will be on the right of hi reervation price. Meanwhile, the buyer target price will be on the left of hi reervation price. In the online motorcycle taxi context, the driver can be repreented a the merchant while the paenger can be repreented a 6137

5 the buyer. So, the negotiation illutration between driver and paenger i hown in Figure 2. Figure 2. Negotiation Area Between Driver and Paenger Negotiation can end with agreement or not. There i rule which negotiation will end with agreement [11]. Negotiation will end with agreement if the left party reervation price i on the right of the right party reervation price [11]. The agreement ucce i not affected by the target point poition. Target point affect in the opening price and the negotiation duration. Let back to the previou example. At the beginning, the firt driver will open propoed price at 4,000 rupiah per kilometer while the econd driver will open propoed price at 3,500 rupiah per kilometer. During the negotiation proce, the driver propoed price will get lower. In the other ide, the paenger will open propoed price at 1,500 rupiah per kilometer. During the negotiation proce, the paenger propoed price will get higher. At the certain time, the paenger propoed price will cro the firt driver propoed price o that the agreement will occur. Meanwhile, for any time, the paenger propoed price will never cro the econd driver propoed price o that the agreement will never occur. The other form of negotiation i auction. Auction i a negotiation between eller and buyer or buyer and eller which one party propoe opening price while other partie compete to cloe the deal by propoing or bidding competitive price. The opening price i uually the reervation price. When the bidding equence run, the end price may exceed other party target point. The paengerdriver auction illutration i hown in Figure 3. The example i a follow. Suppoe that there i paenger whoe target point i 1,500 rupiah per kilometer and reervation price i 2,500 rupiah per kilometer. Now, there are two driver. The firt driver ha target point at 4,000 rupiah per kilometer and reervation price at 2,000 rupiah per kilometer. The econd driver ha target point at 3,500 rupiah per kilometer and reervation price at 3,000 rupiah per kilometer. Baed on thi ituation, negotiation between the paenger and the firt driver will end with agreement while negotiation between the paenger and the econd driver will end with no agreement. The rationale of the example above i a follow. In the negotiation between the paenger and the firt driver, the paenger reervation price i on the right of the firt driver reervation price. So, there i interection area between the negotiating partie. In the negotiation between the paenger and the econd driver, the paenger reervation price i on the left of the econd driver reervation price. So, there i not any interection area between the negotiating partie. Figure 3. Paenger-Driver Auction 6138

6 A it i mentioned in negotiation proce, auction may end with agreement or no agreement. Agreement will reach if there i minimum one bidder who propoe price. The lat bidder who propoe price will win the auction. In other hand, if there are not any bidder who propoe price then the auction will end with no agreement. The main common type of auction i ealed bid auction. In thi form, bidder ubmit hi propoal without knowing other bidder propoal [12]. Beide thi term, baed on the winner, auction can be divided into two type: firt price auction and econd price auction [12]. In the firt price auction, the highet bidder become the auction winner and the deal price i the price that i propoed by him [12]. Meanwhile, the econd price auction i imilar to the firt auction winner. But, the price that mut be paid i the price that i propoed by the econd highet winner [12]. Klemperer explained that there are four baic type of auction: acending bid auction, decending bid auction, firt price ealed bid auction, econd price ealed bid auction [13]. The acending bid auction i alo called a Englih auction [13]. Decending bid auction i alo called a Netherland or Dutch auction [13]. The lat two type of auction are alo called a Vickrey auction [13]. There are everal other auction model. The firt model i one party propoe the opening price. Then, other partie who have willingne to bid will propoe price. The bet propoed price will win the auction. Bidder cannot reubmit price. Thi i ingle round auction. The example i a follow. Suppoe that there i paenger that create travel order. He open the price at 1,000 rupiah per kilometer. Then, there are five driver {m 1, m 2, m 3, m 4, m 5 } that receive the travel order and the propoed price. Thee driver then ubmit their offering price at {3,000; 2,500; 1,700; 1,800; 2,100} rupiah per kilometer conecutively. Becaue thi i a ingle round auction, then the paenger ha only two option: reject or accept. In thi cae, the lowet driver offering price i 1,700 and it i belong to driver m 3. If the paenger reervation price i equal or higher than the lowet offering price, for example it i at 2,500 rupiah per kilometer, the paenger will accept the propoal and driver m 3 will win the auction. Meanwhile, if the paenger reervation price i lower than the lowet offering price, for example it i at 1,500 rupiah per kilometer, the paenger will reject the propoal and the auction will end with no agreement. The ituation i different if thi i a multi round auction. A it i mentioned above, uppoe that in the firt bidding round, the lowet propoing price i 1,700 rupiah per kilometer. Suppoe that the econd bidding round i now open. The next propoed price hould be lower than 1,700 rupiah per kilometer. Becaue of driver m 1 and m 2 reervation price i above 2,000 rupiah per kilometer then only driver m 3, m 4, and m 5 ubmit new offering price in the econd bidding eion. Thee new propoed price are {1,600; 1,500; 1,400} conecutively. Now, the lowet propoed price i at 1,400 rupiah per kilometer and it i belong to driver m 5. Then, the third bidding round i open. Becaue the m 4 and m 5 reervation price i at 1,500 rupiah per kilometer then thee driver do not ubmit the offering price. Driver m 4 alo do not ubmit the offering price becaue hi reervation price i at 1,400 rupiah per kilometer. Becaue there are not any driver who ubmit the propoal, then the bidding eion i end. Becaue the lowet offering price i at 1,400 rupiah per kilometer and thi price i belong to driver m 5 then thi driver become the auction winner at 1,400 rupiah per kilometer price level. 4. PROPOSED MODEL Baed on that condition, in thi paper, we propoe new online motorcycle taxi dipatch model baed on auction model. The propoed model i developed baed on two noveltie o that thi reearch i the improvement of the previou model. The firt novelty i that thi model adopt multi agent ytem which in thi model, the driver and paenger interet are accommodated. In the previou agent baed model, parameter that are concerned are the afety and driver bravery [5]. Meanwhile, in thi reearch, parameter that are included in the multi agent ytem i paenger maximum waiting time, driver maximum pickup and travel ditance, and price level. The econd novelty of thi reearch i the adoption of the auction method. Auction baed model ha not been adopted in any reearche in automatic online taxi dipatch ytem yet. It i becaue many taxi dipatch model are dominated by nearet driver model [7,14,15] or FIFO model [16,17]. 6139

7 In thi ytem, there are two agent type: driver and paenger. There are three model in thi reearch. In the firt model, parameter in the auction i the number i the maximum pickup ditance. In the econd model, parameter in the auction i the number of maximum travel ditance. In the third model, parameter in the auction i the combination between maximum pickup ditance and maximum travel ditance. Thee pickup and travel ditance are the driver interet. Meanwhile, the paenger interet that are accommodated in thee model are waiting time (t wait ) and price. The paenger waiting time i calculated by uing Equation 1. In the Equation 1, the d pickup i the pickup ditance and the v i the vehicle peed. Baed on the paenger view, it doe not matter how far the driver current poition from the pickup location a long a the waiting time i till equal or under the paenger maximum waiting time (t maxwait ). So, it i contradicted with the previou work which ued obervation range to limit pickup ditance. d pickup twait (1) v There are two tep that are required in all propoed model in thi paper. The firt tep i finding driver candidate. The econd tep i dipatching the order to the candidate. Thi proce i executed equentially. The main algorithm of thi proce i hown in Figure 4. Begin C find_candidate(m) if n(c) > 0 then m el dipatch(c) ele fail end Figure 4. Main Proce Algorithm In the main proce algorithm, ome variable and function are ued. Variable C repreent the et of driver candidate o that the n(c) i the number of the C et member. Variable M repreent the et of driver in the ytem. Variable m el repreent the driver that i elected to execute the order. The find_candidate function repreent the firt tep. The dipatch function repreent the econd tep. It i hown that the number of C et member i more than 0 then the proce continue to the econd tep. Otherwie, the ytem fail to find candidate. The difference between the firt, the econd, and the third propoed model are in the candidate finding procee. There are three requirement for every model o that driver in the ytem can be a candidate: price requirement, waiting time requirement, and ditance requirement. But, there are imilaritie imilaritie among model. The imilaritie are the reervation price requirement and waiting time requirement. In all model, the driver reervation price mut be lower than or equal to the paenger reervation price. Thi requirement i repreented in Equation 2. In Equation 2, variable price repreent the price requirement tatu. price 1, mre p 0, ele re (2) Beide price requirement tatu, the waiting time requirement tatu mut be calculated too. The price requirement tatu i determined by uing Equation 3. In Equation 3, variable wait repreent the waiting time requirement tatu. Meanwhile, the t maxwait repreent the maximum waiting time. So, baed on thi formula, the waiting time requirement tatu will be 1 only if the waiting time i equal to or lower than the maximum waiting time. wait 1, twait tmax 0, ele wait (3) Beide price requirement and waiting time requirement tatue, the ditance requirement tatu mut be calculated too. The ditance requirement tatu i repreented by uing variable ditance. The driver will become a driver candidate for the order if thi driver i available and hi both price requirement tatu and ditance requirement tatu i 1. Thi proce i repreented in candidate finding algorithm that i hown in Figure 5. The explanation of the candidate finding algorithm i a follow. For the firt time, the candidate et i cleared. Then, the proce iterate from the firt driver to the lat driver in the ytem. The driver availability tatu i repreented by uing variable av. If the driver i available then the tatu i 0. Otherwie the tatu i 1. The calc_price function i ued for calculating the price requirement tatu. The calc_ditance function i ued for calculating the ditance requirement tatu. The calc_wait function i ued for calculating the 6140

8 waiting time requirement tatu. All tatue then are ummed and the reult i tored in variable tot. If all requirement tatue are 1 then thi driver will added to the candidate et. So, if there i not any driver who join into the candidate et then the number of it member will be zero. begin C.clear() for i = 1 to n(m) begin if av,i = 0 then begin price,i calc_price(m i ) ditance,i calc_ditance(m i ) wait,i calc_wait(m i ) tot,i price,i + ditance,i if tot,i = 3 then C.addmember(m i ) end end end Figure 5. Candidate Finding Algorithm In the firt propoed model, the ditance parameter that i calculated i the driver pickup ditance. In thi model, the driver pickup ditance mut be equal to or lower than hi maximum pickup ditance. The ditance requirement tatu i determined by uing Equation 4 to Equation 6. d pickup pickup p p (4), d pickup m d 1 pickup max pickup 0, ele (5) di tan ce pickup (6) The explanation of thee equation i a follow. In Equation 4, the pickup ditance (d pickup ) i the Euclidean ditance between pickup location (p pickup ) and driver current location (p m ). Then, in Equation 5, the pickup tatu will be 1 only if the pickup ditance i equal to or lower than the driver maximum pickup ditance (d maxpickup ). Otherwie, the pickup tatu i 0. In Equation 6, it i hown that in thi model, the ditance requirement tatu i the pickup tatu. In the econd propoed model, the ditance requirement tatu i the travel ditance. The travel ditance mut be equal to or lower than the driver maximum pickup ditance. The ditance requirement tatu of the econd propoed model i determined by uing Equation 7 to Equation 9. d travel travel p p (7) det, d pickup d 1 travel max travel 0, ele (8) di tan ce travel (9) The explanation of thee equation i a follow. In Equation 7, the travel ditance (d travel ) i the Euclidean ditance between detination location (p det ) and pickup location (p pickup ). Then, in Equation 8, the travel tatu will be 1 only if the travel ditance i equal to or lower than the driver maximum travel ditance (d maxtravel ). Otherwie, the travel tatu i 0. In Equation 9, it i hown that in thi model, the ditance requirement tatu i the travel tatu. In the third propoed model, the ditance requirement tatu i calculated baed on the pickup tatu and the travel tatu. All of thee tatue mut be 1. Otherwie the ditance requirement tatu will be 0. The pickup tatu i determined by uing Equation 5. The travel tatu i determined by uing Equation 6. Meanwhile, the ditance requirement tatu i determined by uing Equation 10. di 1, pickup 1 travel 1 tan ce (10) 0, ele After finding the candidate, the next tep i allocating the order to the certain driver. In thi tep, auction method i implemented. In our propoed model, the type i ealed ingle round auction. It mean that driver can propoe once and cannot reubmit price propoal. The benefit of thi method i the auction proce will be impler and fater. But, the driver ha only one chance to ubmit propoal. Thi dipatch tep i divided into two ub tep: propoal ubmiion and price finalization. Thee ub tep are done equentially. In the propoal ubmiion, every candidate will end price propoal. Then, the ytem will decide the auction winner. After the winner i decided, the next tep i determining the final price. The price that i propoed by the winner mut be adjuted with the paenger interet. In propoal ubmiion tep, each driver will end hi price propoal. In the auction model, we ue variable p a price even we have ued 6141

9 variable p a location in the previou tep. The propoed price that i ubmitted by the driver i repreented by uing variable p prop. The winner of auction i the driver who propoe the lowet price. The formula i decribed in Equation 11 and the proce i hown in Figure 6. m el p prop m m C min (11) begin curel 1 p min p(m curel ) for j = 2 to n(c) begin if p(m j ) < p min then begin p min p(m j ) curel j end end m el m curel end Figure 6. Winner Deciion Proce Algorithm The explanation of the algorithm i a follow. Variable curel tore the index of the current elected merchant. Variable p min tore the current minimum price. So, at the beginning, the current elected merchant i the firt merchant in et C. Then, there i looping proce from index 2 to the number of et C. For each iteration, the proce check whether the the current indexed driver propoe lower price. If the current indexed driver propoe lower price, then he will be the current elected driver and hi price will be the current lowet price. After the looping i ended, the current elected driver will be the elected driver. After the elected driver i determined, the next tep i finalizing the price. Thi driver propoed price may be higher than the paenger reervation price. To make it fair, the final price mut accommodate both partie: driver and paenger. In thi model, the final price i in the middle between the driver reervation price and the paenger reervation price. Thi formula i decribed in Equation 12 and Equation 13. pre p final pre _ driver (12) 2 p p p (13) re re _ paenger re _ driver There are new variable in Equation 12 and Equation 13. Variable p final i the final price. Variable p re_driver i the driver reervation price. Variable p re_paenger i the paenger reervation price. Variable p re i the difference between paenger reervation price and driver reervation price. Baed on Equation 12 and Equation 13, it i hoped that there i fairne in final price. When the paenger reervation price i higher than driver reervation price, the final price will be higher than the driver reervation price and lower than the paenger reervation price. Meanwhile, when the driver reervation price i equal to the paenger reervation price then the final price i at both reervation price. 5. IMPLEMENTATION Thee three propoed model then i implemented into online motorcycle taxi dipatch imulation application. The world of the imulation i a virtual city. It ize i 20 kilometer width and 20 kilometer length. Paenger and driver are ditributed uniformly in the city. The application i not time variant imulation. The imulation cenario i a follow. At the beginning, ome paenger and driver are generated in the application. All parameter that are related to thee entitie are generated too. Parameter that are related to the paenger are: maximum waiting time, reervation price, target point, pickup location, and detination location. Parameter that are related to the driver are: current location, maximum pickup ditance, maximum travel ditance, peed, reervation price, and target point. Thee parameter are generated randomly. After all of thee parameter are et then the dipatch proce run. The dipatch proce iterate from the firt to the lat order. If the dipatch proce i ucce then the order tatu i et ucce. Otherwie, the order tatu i et fail. In the end of the application, the ucce ratio (r ucce ), which i the ratio between the number of ucce order and the number of total order are calculated. Beide that, the other oberved parameter are: driver total revenue, paenger total waiting time, and driver total pickup ditance. 6. DISCUSSION After the propoed model i implemented into imulation application, then thee three propoed model are teted to evaluate their 6142

10 performance. Thee propoed model are compared to each other. Beide, the previou nearet driver model i teted too o that the performance of thee propoed model i alo compared with the previou model [6]. The imulation evaluate the relation between the increaing number of paenger and the oberved parameter. In thi imulation, the number of driver i et 500 driver. The number of paenger i et from 10 to 100 paenger. The adjuted parameter are et default. The default value of thee adjuted parameter i hown in Table 5. In previou model, the driver price i 2,000 rupiah per kilometer. The firt reult i the driver ucce ratio. A it i mentioned above, the ucce ratio i the ratio between the number of ucce order and the number of total order. The reult i hown in Table 6. The reult i collected from the firt model, the econd model, the third model, and the previou nearet driver model. To make it i oberved eaier, the data trend i alo hown in Figure 7. Table 5. Adjuted Parameter Default Value Parameter t maxwait p re_paenger p target_paenger d max_pickup d max_travel v p re_driver p target_driver Default Value 10 minute 4,000 rupiah/km rupiah/km 3 km 12 km 0.5 km/minute 1,500 rupiah/km 5,000 rupiah/km Table 6. Driver Succe Ratio Reult n p (unit) Firt Model Second Model Third Model Nearet Driver Model driver to execute their order. When the number of paenger increae, the driver ucce ratio decreae. Thi condition i different when dipatch ytem ue previou nearet driver model [6]. By uing thi model, the driver ucce ratio i alway 100 percent. It mean that all of paenger order are executed uccefully. Figure 7. Succe Ratio Baed on data in Table 6, it i hown that by uing all of the propoed model, the driver ucce ratio never reache 100 percent. It mean that there are ome paenger that cannot get the The reaon i a follow. When the ytem implement thee propoed model, there are many interet that mut be accommodated. So, when thee interet are not accommodated, there are not any driver will execute order. The condition i different with the nearet driver model [6]. Becaue there i not any ditance limitation, a long a there i available driver in the ytem, the order will be executed even the pickup ditance i very far. Baed on data in Figure 7, the comparion among propoed model in driver ucce ratio i a follow. The econd model perform the highet 6143

11 ucce ratio. The third model perform the lowet ucce ratio. The firt model perform in the middle. When the number of paenger i low, the ucce ratio of the firt model i cloe to the ucce ratio of the econd model. When number of paenger increae, the ucce ratio of the firt model decline fater than other propoed model. When the number of paenger i high, the ucce ratio of the firt model i cloe to the ucce ratio of the third model. The econd reult i driver total revenue. The reult i hown in Table 7. The reult i collected from the firt model, econd model, third model, and the previou nearet driver model. The data trend i alo hown in Figure 8. Table 7: Total Driver Revenue Reult n p (unit) Firt Model (rupiah) Second Model (rupiah) Third Model (rupiah) Nearet Driver Model (rupiah) , , , , , , , , , , , , ,585 1,210, ,185 1,313, ,005 1,339, ,415 1,588, ,275 1,376, ,720 1,890, ,160 1,709, ,785 2,205, ,050 1,910, ,345 2,491, ,470 1,989, ,085 2,862, ,240 2,304, ,075 3,150,000 the revenue of the econd model i higher than the revenue of the firt model. Thi condition i the conequence of the ucce ratio. Comparing to each other, it i hown that driver ucce ratio ha poitive correlation with the total driver revenue. Baed on data in Figure 8, it i hown that at the beginning, the revenue gap among model i narrow. During the increaing of the number of paenger, the revenue gap among model i wider. It i becaue during the increaing of the number of paenger, the ucce ratio gap i wider too and it caue the wider gap in total driver revenue. Figure 8. Driver Revenue Baed on data in Table 7, it i hown that the ytem produce the highet total revenue when it implement nearet driver model. In the other ide, the ytem produce the lowet total revenue when it implement the third propoed model. Meanwhile, the revenue that i generated by the firt and the econd model are between them with Even thee propoed model produce lower total revenue than the previou model doe, the average revenue that i generated by uing thee propoed model i higher than by uing the previou model [6]. Let take example when the number of paenger i 100 peron. The dividing the total revenue with the ucce ratio, the reult i a follow. The firt model average revenue i 52,192 rupiah per order. The econd model average revenue i 39,332 rupiah per order. The third model average revenue i 39,195 rupiah per order. The previou model [6] average revenue i 31,500 rupiah per order. 6144

12 Baed on thi data, the firt model produce the highet average driver revenue among all model. Then, it i followed by the econd and the third model. The average revenue gap between the econd and the third model i very tight. Meanwhile, the previou nearet driver model produce the lowet average driver revenue. The reaon of thi condition i a follow. In the firt model, the travel ditance i not limited. So, driver can get high travel ditance order. Thi condition i different to the econd and the third model. In thee model, the travel ditance limitation i applied. So, only medium or low travel ditance order that can be allocated to the driver. Meanwhile, the condition in the previou nearet driver model i different. In thi model, fix tariff i applied and all order get ame price at 2,000 rupiah per kilometer. Thi price i cloe to the driver reervation price. The third reult i paenger total waiting time. The reult i hown in Table 8. The reult i collected from the firt model, econd model, third model, and the previou nearet driver model. The data trend i alo hown in Figure 9. Table 8: Total Paenger Waiting Time n p (unit) Firt Model (minute) Second Model (minute) Third Model (minute) Nearet Driver Model (minute) , , , , ,078 The firt and the third model are in the middle with the waiting time of the firt model i higher than waiting time of the third model. Thi condition i related to the driver ucce ratio. The ucce ratio ha poitive correlation with the total paenger waiting time. Baed on data in Figure 9, it i hown that the number of paenger affect the waiting time gap among model. When the number of paenger i low, the gap between the highet and the lowet waiting time i narrow. When the number of paenger increae, the gap i wider. Figure 9. Paenger Waiting Time Baed on data in Table 8, it i hown that in total paenger waiting time apect, the nearet driver model perform the highet one. Meanwhile, the third propoed model perform the lowet one. The next analyze i the average paenger waiting time. The example i the waiting time when the number of paenger i 100 peron. The average waiting time i gotten by dividing the total waiting time with the ucce ratio. The average waiting time of the firt model i 10.6 minute. The average waiting time of the econd model i 11 minute. The average waiting time of the third model i 9.4 minute. The average waiting time of the fourth model i 20.8 minute. 6145

13 Baed on thi reult, it i hown that when the ytem implement nearet driver model, the average waiting time i the highet. The lowet reult i reached when the ytem implement the third model. The average waiting time of the firt and the econd model are in the middle with the firt model perform lower than the econd model doe. The fourth reult i driver total pickup ditance. The reult i hown in Table 9. The reult i collected from the firt model, econd model, third model, and the previou nearet driver model. The data trend i alo hown in Figure 10. Table 9:Driver Total Pickup Ditance Reult n p (unit) Firt Model Second Model Third Model Nearet Driver Model A it i hown in Figure 10, the correlation between the number of paenger and the total pickup ditance i depended on the model that i implemented. When the ytem implement the econd model or the previou model, the number of paenger ha poitive correlation with the total pickup ditance with ignificant influence. Meanwhile, when the ytem implement the firt model or the third model, the number of paenger ha le ignificant influence to the total pickup ditance. Figure 10. Pickup Ditance Baed on data in Table 9, it i hown that in total pickup ditance apect, the econd model perform the highet one. Meanwhile, the third model perform the lowet one. The previou model perform lower than the econd model doe but the gap between them i cloe. The firt model perform higher than the third model but the gap between them i low. The next analyze i the average pickup ditance. The example i the average pickup ditance when the number of paenger i 100 peron. The reult i gotten by dividing the total pickup ditance with the ucce ratio. The average pickup ditance of the firt model i 4.3 kilometer. The average pickup ditance of the econd model i 10.1 kilometer. The average pickup ditance of the third model i 3.9 kilometer. The average pickup ditance of the previou model i 5.82 kilometer. Baed on thi reult, it i hown that the third model perform the lowet average pickup ditance. Thi performance i followed cloely by the firt model. Then, the econd model perform the wort pickup ditance. Meanwhile, the previou model [6] perform moderate. 6146

14 7. CONCLUSION AND FUTURE WORK Baed on the explanation above, the propoed model ha been developed and implemented into online motorcycle taxi dipatch imulation. The model i developed baed on auction method and accommodate both driver interet and paenger interet. The interet that are accommodated in thi model are: reervation price, target point, waiting time, pickup ditance, and travel ditance. In thi reearch, we propoe three model. The firt model accommodate the pickup ditance interet. The econd model accommodate the travel ditance interet. The third model accommodate both pickup ditance and travel ditance. In all model, the price range and waiting time are accommodated. In all model, the ealed ingle round auction implemented where the winner i driver meet all requirement and ubmit the lowet price. Thee propoed model then being teted to evaluate their performance. The oberved variable are: ucce ratio, revenue, waiting time, and pickup ditance. Performance of thee model i alo compared with performance of the previou nearet driver model. Related to the hypothei that i mentioned in the firt chapter, the concluion i a follow. Among all propoed model, ome model perform better than the previou nearet driver model while other model perform wore. In ome parameter, the propoed model perform better than the previou nearet driver model while in other parameter, the propoed model perform wore. The detailed concluion i explained in the reearch finding below. The reearch finding i a follow. In ucce ratio apect, the nearet driver model perform the bet and it i followed by the econd model, the firt model, and the third model conecutively. In average driver revenue apect, the firt model perform the highet average driver revenue and it i followed by the econd model, the third model, and the previou model conecutively. In average waiting time apect, the third model perform the lowet average waiting time apect and it i followed by the firt model, the econd model, and the previou model conecutively. In average pickup ditance apect, the third model perform the lowet average pickup ditance and it i followed by the firt model, the previou model, and the econd model. Thi reearch trigger other reearch potential. A the online motorcycle taxi buine i till growing, it buine model i till improved. Dipatch mechanim i more complex too becaue takeholder need are more complex too. Related to thi reearch, the auction model that i implemented in thi model i not the only auction model. For example, multi round auction baed mode will be very intereting to be implemented o that the performance between auction model can be compared. In a broader view, auction i mall part of negotiation model. Implementing other negotiation model in the online motorcycle taxi dipatch ytem i alo challenging. REFERENCES: [1] P. Mukita, What Doe The Grab-Uber Deal Mean For Go-Jek, Techinaia, March 28 th, 2018, [2] L. Coeboom, GrabBike and Go-Jek Prepare for a Street Fight in Indoneia, Techinaia, June 9 th, 2015, indoneia-grabbike-go-jek-competition. [3] A. Aravindan and H. Somerville, Uber Sell Southeat Aia Buine to Grab After Cotly Battle, Reuter, March 26 th, 2018, uber-ell-outheat-aia-buine-to-grab-after -cotly-battle-iduskbn1h204k. [4] J. Kollewe, Uber to Sell South-eat Aia Buine to Competitor Grab, The Guardian, March 26 th, 2018, com/technology/2018/mar/26/uber-to-ellouth-eat-aia-buine-to-competitor-grab. [5] P.D. Kuuma, Online Motorcycle Taxi Simulation by Uing Multi Agent Sytem, International Journal of Applied Engineering Reearch, vol 19(12), 2017, pp [6] P.D. Kuuma, Nearet Driver-FIFO Combination Model in Online Motorcycle Taxi Dipatch Sytem, Journal of Theoretical and Applied Information Technology, vol 95(19), [7] K.T. Seow, N.H. Dang, D.H. Lee, A Collaborative Multiagent Taxi-Dipatch Sytem, IEEE Tranacntion on Automation Science and Engineering, vol. 7(3), 2010, pp [8] A. Kim, M.E. Lewi, C.C. White, Optimal Vehicle Routing with Real-time Traffic Information, IEEE Tranaction on Intelligent Tranportation Sytem, vol 6(2), 2005, pp

15 [9] H. Raiffa, The Art and Science of Negotiation, Cambridge, Harvard Univerity Pre, [10] R. Fiher, W. Ury, B. Patton, Getting to Ye: Negotiating an Agreement Without Giving In, Random Houe Buine Book, [11] G.N. Herman, J.M. Cary, J.E. Kennedy, Legal Couneling and Negotiating: A Practical Approach, Matthew Bender and Company, [12] L.M. Auubel, Auction: Theory For The New Palgrave 2 nd Edition, Univerity of Maryland, [13] P. Klemperer, A Survey of Auction Theory, Oxford Univerity, [14] Z. Liao, Taxi Dipatching via Global Poitioning Sytem, IEEE Tranaction on Engineering Management, vol 48(3), 2001, pp [15] F. Miao, S. Lin, S. Munir, J.A. Stankovic, H. Huang, D. Zhang, T. He, G.J. Pappa, Taxi Dipatch with Real Time Sening Data in Metropolitan Area: A Receding Horizon Control Approach, IEEE Tranaction on Automation Science and Engineering, vol 13(2), [16] Y. Lu, S. Xiang, W. Lu, Taxi Queue, Paenger Queue or No Queue? A Queue Detection and Analyi Sytem Uing Taxi State Tranition, Proceeding of 18 th International Conference on Extending Databae Technology, 2015, March 23-27, Bruel. [17] D. Ang, J. Shenfeng, L. Guanhua, W. Wei, Improving Taxi Boarding Efficiency at Changi Airport, Singapore Univerity of Technology and Deign, report. 6148