Pick-up point recommendation strategy based on user incentive mechanism

Motivation scenario

The spatial crowdsourcing task model under the ride-hailing platform is shown in Fig. 4. Users submit travel task requests in real time from their mobile devices, with each task specifying a location, budget, and deadline. After receiving the task request from the user, according to the driver’s location, the platform matches according to certain matching rules if the matching result meets the constraints of the current task. The platform generally selects appropriate workers to be assigned to different users from the perspective of improving the probability of completing tasks as much as possible, maximizing the number of tasks completed or minimizing the cost of tasks, etc. Finally, both the driver and the user are notified of the current matching results from the platform. The driver needs to arrive at the user’s location to complete the pick-up service before the deadline of the task request, and then send the user to the destination to complete the execution of the crowdsourcing task.

Problem scenario

After the platform matches the user with the driver, the driver needs to go to the user’s current location. There may be some problems here. For example, Fig. 5 describes a ride-hailing task pick-up scenario.

In Fig. 5, user u₁ initiates a car-hailing service at the current location and matches the nearest driver w₁ according to the nearest policy, but the current driver w₁ is already at the intersection or in the right turn, straight go or left turn lane. When the driver w₁ is in the straight go lane, it can quickly receive the user as shown in Fig. 5A. The driver needs to follow the current lane, make a U-turn at the next intersection, and return to the intersection to catch the user, as shown in Figs. 5B and 5C. The appeal case illustrates that fixing the user’s location as the pick-up point may greatly increase the execution cost of the task in some cases, so selecting the right pick-up point is an urgent problem to be solved. The goal of this article is to optimize the total cost as an indicator. An incentive mechanism is proposed to encourage users select the right pick-up point, so as to reduce the overall travel cost of the order and improve the travel efficiency.

Problem definition

Ride-hailing platform can be regarded as a spatial crowdsourcing platform. Based on current literature and practical application scenarios, this section will introduce some definitions and concepts related to the problem.

Definition 1 (Worker, w). The spatial crowdsourcing platform includes a set of crowdsourcing workers W = {w₁, w₂, …, w_n}, worker w ∈ W are the recipients and performers of the crowdsourcing tasks. They receive and complete the crowdsourcing tasks after registration on the platform. The workers will report their current location locw_i = (lw_i(x), lw_i(y)) to the platform, lw_i(x), lw_i(y) respectively indicating the longitude and latitude of the current location of the worker. Each worker is associated with a set of attributes, which are expressed as w_i = {locw_i, rw_i, v_i}, where rw_i represents the crowdsourcing tasks accepted by the current worker w_i or assigned by the platform, and v_i is the moving speed of the worker on the road section. In the model, the worker accepts the task assigned by the platform. The preference of the worker is not considered in the task assignment process, and the platform assigns tasks to the worker according to the principle of maximizing the global objective function. Each worker can only match one task at a time.

Definition 2 (Worker, u). The spatial crowdsourcing platform includes a set of users U = {u₁, u₂, …, u_m}, which are the publishers of spatial crowdsourcing tasks, expressed as u ∈ U. They publish crowdsourced tasks after registration on the platform. When they publish crowdsourced tasks, users will report their current location locu_j = (lu_j(x), lu_j(y)) to the platform. Each user is associated with a set of attributes u_j = {locu_j, walk_j, ru_j}, where walk_j represent the maximum distance that the current user is willing to walk to the surrounding pick-up points, and ru_j represent the crowdsourced tasks published by user u_j.

Definition 3 (Task, r). Spatial crowdsourcing tasks are published by users on the spatial crowdsourcing platform, and each spatial crowdsourcing task usually carries time attributes and geographical location information. The tasks of the online ride-hailing platform are divided into two parts, one is the pre-pick-up task that occurs before the start of the journey, and the other is the journey task after the completion of the pick-up. Represents as r_k = {t_k, ori_k, des_k, pri_k, pick_k}, where r_k represents the release time of task k, ori_k, des_k represents the original starting point and destination point location of the task k respectively, pri_k isthe order amount of the task, pick_k is the final determined pick-up point location, and the initial value is locu_j.

Definition 4 (Pick-up point, p). For a given task r_k, the platform calculates the suitable pick-up point location of the task according to the algorithm combined with the information given. The pick-up point is selected from the road node or interest point, and both user and worker need to go to the selected pick-up point.

Definition 5 (Task execution cost, C). The completion of each task will have execution cost, and the order cost consists of two parts: distance cost and time cost. Firstly, distance cost is the sum of the distance cost from the pick-up point to the destination and the distance cost of the driver from the current location to the pick-up point, which is shown as Eq. (1). Secondly, time cost is the sum of the user’s waiting time cost and pick-up time cost. As shown in Eq. (2). dist(O, D) represents the road network distance from the origin point O to the destination point D , which takes into account the driving direction and turning information of the actual road section, rather than directly using the European distance or Manhattan distance, and can better reflect the true cost of task execution. The total cost of task execution is the sum of distance cost and time cost, which is shown as Eq. (3). (1)${\displaystyle C_{r_k}^d\left(w_i,u_j\right)=dist\left(loc{w}_i,or{i}_k\right)+dist\left(loc{u}_j,or{i}_k\right)+dist\left(or{i}_k,de{s}_k\right)}$ (2)${\displaystyle C_{r_k}^t\left(w_i,u_j\right)=\frac{dist\left(pic{k}_k,de{s}_k\right)}{v_i}}$ (3)${\displaystyle C_{r_k}\left(w_i,u_j\right)=C_{r_k}^d\left(w_i,u_j\right)+C_{r_k}^t\left(w_i,u_j\right)}$ where $C_{r_k}^d\left(w_i,u_j\right)$ represents the execution distance cost when a crowdsourced task posted by a user is assigned to a worker. $C_{r_k}^t\left(w_i,u_j\right)$ is the corresponding execution time cost. C_rk(w_i, u_j) represents the total execution cost of the task, kd and kt are distance weight and time weight respectively.

As shown in Eq. (3), if the selected pick-up point is close to both the user and the driver, and the destination can be reached in a convenient way, the task cost will be relatively low. The execution cost of the overall task is calculated by summing the total cost of all tasks, as shown in Eq. (4). (4)${\displaystyle C_r=\sum _{r_k\in R}{C}_{r_k}\left(w_i,u_j.\right)}$

Definition 6 (Incentive mechanism). In the general spatial crowdsourcing platform, it is always assumed that workers move to the task release point, which will cause detour problems on the real road network. In order to reduce the task execution cost and user waiting time, this article proposes an incentive mechanism to encourage users to walk to the appropriate pick-up point within a certain range, and the reward is used to offset the trip price, which is shown as Eq. (5). Using this as an incentive for users does not require the platform to bear additional costs, but is directly reflected in the reduction of the price of travel orders, which not only reduces the cost of taking the driver, but also reduces the travel consumption for users. (5)${\displaystyle R=P\left(dist\left(loc{u}_i,de{s}_k\right)\right)-P\left(dist\left(pci{k}_k,de{s}_k\right)\right)}$ where P(dist(O, D)) represents the travel amount under the road network distance from the origin O point to the destination point D, which is shown as Eq. (6). (6)${\displaystyle P\left(dist\left(O,D\right)\right)=10+dist\left(O,D\right)\ast 3.}$

Definition 7 (User walking incentive benefit ratio, Rw). In the case that there are multiple pick-up points within a given range, the walking distance of different pick-up points is different, and the execution cost of each pick-up point is also different. Therefore, this article defines the walk incentive benefit ratio to determine the walk benefit of users at each pick-up point, which is shown as Eq. (7). Then the user can choose the pick-up point with high walking incentive income. (7)${\displaystyle Rw\left(p_{r_k}^l\left(u_j\right)\right)=\frac{dist\left(loc{u}_j,de{s}_k\right)-dist\left(pci{k}_k,de{s}_k\right)}{dist\left(loc{u}_j,pci{k}_k\right)}}$ where $p_{r_k}^l\left(u_j\right)$ represents the lth pick-up point under the user’s maximum walking range walk_j in the case of the crowdsourcing task r_k released by user u_j, Rw is the user’s walking benefit ratio of the pick-up point. It can be seen from the above equation that the greater the distance difference between the pick-up point and the destination point compared with the original starting point and the distance between the user and the pick-up point, the greater the walking benefit of the pick-up point to the user.

Definition 8 (Driver benefit ratio, Rd). Accordingly, for drivers, the driving costs saved by different driving points are also different, including two driving costs, one is the driving cost of the driver’s current location to the driving point, and the other is the driving cost of the driver to the destination from the driving point. Suitable selection of driving points can save the costs of the above two driving stages, so this article defines the driver driving benefit ratio. As shown in Eq. (8), the driving cost and benefit of each driving point are determined. (8)${\displaystyle Rd\left(p_{r_k}^l\left(w_i,u_j\right)\right)=\frac{dist\left(loc{w}_i,loc{u}_j\right)+dist\left(loc{u}_j,de{s}_k\right)}{dist\left(loc{w}_i,pci{k}_k\right)+dist\left(pci{k}_k,de{s}_k\right)}}$ where $p_{r_k}^l\left(w_i,u_j\right)$ means that the crowdsourcing task published by the user u_j is assigned to worker w_i, and the driving benefit ratio of the driver at the lth pick-up point under the maximum walking range walk_j of the user. Rd is the driving benefit ratio of the driver at the pick-up point. It can be seen from the Eq. (8) that the greater the ratio between the original distance of the task and the distance through the pick-up point, the greater the driving benefit of the pick-up point to the driver.

Definition 9 (Forward rate, Rf). By encouraging the user to walk the appropriate distance to the appropriate pick-up point within the maximum walking distance of the user, the driving distance of the driver can be effectively reduced, and the driving cost of the order execution can be reduced, including the driving cost of the pick-up distance and the driving cost of the drop-off distance in the first stage of the task. However, in the real road network environment, with the increase of the user’s maximum walking distance, the number of potential candidate pick-up points around the user increases exponentially. In the case of more pick-up points to choose from, the calculation of each return index of each pick-up point is too large, which cannot meet the needs of real-time interaction scenarios. For this reason, this article defines an indicator of direction rate that includes many factors. As shown in Eq. (9), it can be used to evaluate the score of each pick-up point, the diagram is shown in Fig. 6. (9)${\displaystyle Rf\left(p_{r_k}^l\left(w_i,u_j\right)\right)=\frac{dist\left(loc{w}_i,de{s}_k\right)}{dist\left(pci{k}_k,de{s}_k\right)}+\frac{cos\left(\overrightarrow{pci{k}_k,de{s}_k},\overrightarrow{pci{k}_k,loc{w}_k}\right)}{cos\left(\overrightarrow{loc{u}_i,de{s}_k},\overrightarrow{loc{u}_i,loc{w}_k}\right)}.}$

The formula for calculating the forward rate is shown above, which represents the ratio of the driver’s direct distance to the destination and the distance from the pick-up point to the destination and the sum of the cosine value from the initial position to the destination and the pick-up point position to the destination. Using the ratio of the two angles instead of the road network distance greatly reduces the calculation cost.

Travel characteristics analysis

In order to study the order information of peak travel period, it is necessary to analyze the travel characteristic data. This study takes Didi Chuxing Didi Gaia data 2016 Chengdu as an example. This data includes the local order data of Chengdu Second Ring Road in November 2016, which will also be used as the experimental data of this article.

Based on the all-day travel order information on November 1, 2016, heat maps of travel hotspots are generated. In Fig. 8, it can be analyzed from the spatial characteristic distribution of order distribution that the demand for e-hailing data is mainly concentrated in the downtown area, that is, in the third ring road area of Chengdu, especially in the second ring road area, while the demand for travel in other regions is low.

According to the statistics of the order data by hour, the trend of the total order volume per hour is shown in Fig. 9. At 9 am, the demand for online car booking reaches the peak in the morning peak, followed by the second peak demand between 1 pm and 2 pm, and then the evening peak between 1 pm and 2 pm.

The order data are analyzed according to the processing time as shown in Fig. 10. For long-distance travel orders whose processing time is longer than 1 h, the cluster analysis is carried out at five-minute intervals for orders whose processing time is less than 1 h, and the above figure is obtained. As can be seen from the figure, 79% of the orders are processed within 5 to 30 min. The 15-20, 20-25, 25-30 min orders occupy the top three. Based on the above analysis of travel characteristics, this article extracts orders from hot travel areas during peak hours as the input data of this article.

Driver location algorithm

This section designs a link matching algorithm of anchor points. When an order is assigned to a driver, the location information of the driver needs to be determined. This algorithm is used to generate driver information, ensure that the generated anchor point information is on the road section, and generate the driving speed according to the information of the road section. The pseudocode of driver location algorithm is shown in Algorithm 1.

 
_______________________ 
 Algorithm 1: Driver location algorithm                                      _________ 
    Input:   The task set R. 
   Output:   The worker set W 
 1   for each task rk  in R do 
 2       Initial value w 
 3       while w.lw(x),w.lw(y)  not satisfies judgmentconditions, 
    then 
 4          w.lw(x) = orik(x) + Random(0.1,1); 
 5          w.lw(y) = orik(y) + Random(0.1,1); 
 6       end while 
 7       w.v = Random(40,loadmaxspeed) 
  8       W ← W ∪{w} 
 9   end for 
10   return worker set result W    

Restriction 1: The generated driver location information is within a circle with a radius of 1 km from the center of the user, which is shown as Eq. (10). (10)${\displaystyle dist\left(u_i,w_i\right)=\sqrt{{\left(u_i\left(x\right)-w_i\left(x\right)\right)}^2-{\left(u_i\left(y\right)-w_i\left(y\right)\right)}^2}\le 1}$

Restriction 2: And it must be satisfied that the anchor point is on the road section. That is the distance from the nearest road side is less than 20 m, which is shown as Eq. (11). (11)${\displaystyle dist\left(loc{w}_i,load\right)\le 0.02}$

The initial speed of the driver is set to be less than the maximum speed of the road section, which is shown as Eq. (12). (12)${\displaystyle 40\le w_{iv}\le loa{d}_{maxspeed}.}$

Filter candidate pick-up point

When a user launches a crowdsourcing task, there are two location information, the user origin point O and the destination location D, in addition, the algorithm is designed based on the user incentive mechanism to recommend the user to the appropriate pick-up point, the user’s maximum walking distance is walk_j, and the cost of motivating the user to walk is i. When the user initiates an order, the platform finds a suitable collection of pick-up P_k = {p₁, p₂, …, p_m} points according to the user’s destination and the current user’s location, and calculates the order cost.

The original task execution cost is obtained from Eq. (3). After selecting the optimal pick-up point, the distance cost of the order becomes the sum of the distance cost from the pick-up point to the destination and the distance cost from the current location to the pick-up point. Secondly, the time cost becomes the sum of the time from the user’s current location to the boarding point and the maximum time from the driver to the boarding point and the time from the user to the destination after the boarding.

According to the user’s maximum walking distance and maximum waiting time, the algorithm first selects the candidate pick-up point collection that meets the conditions nearby. The pseudocode of filter candidate pick-up point algorithm is shown in Algorithm 2.

 
________________________________________________________________________________________ 
  Algorithm 2: Filtrate candidate pick-up point algorithm                _________ 
    Input:   The user uj, task rk, road network N 
    Output:   The candidate pick-up point setPk  of task rk 
  1   Initial value Pk ←∅ 
 2   for each point p ∈ N 
 3       if p satisfies judgmentconditions, then 
 4          Pk ← Pk ∪{p} 
 5   end for 
 6   return candidate pick-up point set result Pk    

The candidate pick-up point need to meet the following restrictions:

Restriction 3: The selection of the pick-up point needs to meet the distance from the user is less than the user’s maximum walking distance, which is shown as Eq. (13). (13)${\displaystyle dist\left(u_i,p_{i,j}\right)=\sqrt{{\left(u_i\left(x\right)-p_{i,j}\left(x\right)\right)}^2-{\left(u_i\left(y\right)-p_{i,j}\left(y\right)\right)}^2}\le u_{i,w}}$

Restriction 4: Time limit, the selected pick-up point should ensure that the driver can arrive within the maximum waiting time of the user, which is shown as Eq. (14). (14)${\displaystyle \frac{dist\left(w_k,p_{i,j}\right)}{v_k}\le \left(t_i+w{t}_i-p\right)}$

Recommend pick-up point

 
__________________________________________________________________________________________________________ 
  Algorithm  3: Recommendation  strategy  based  on  user  incentive- 
  greedy (BUIRS-G)                                                                   _________ 
    Input:   The user uj, worker wi  task rk, candidate pick-up 
           point setPk, road network N 
    Output:   The recommend pick-up point setPk  of task rk 
  1   for each point p ∈ Pk 
  2          Cp ← Crkwi,uj,p 
 3          rp ← rewardpuj,rk,p 
 4          Income_wp ← Rw( 
 pprk (uj)) 
 5          Income_dp ← Rw( 
 pprk (wi,uj)) 
 6   end for 
 7   for each point p ∈ Pk 
  8       if p not satisfies judgmentconditions, then 
 9          Pk ← Pk∖{p} 
10   end for 
11   return the recommend pick-up point set result Pk    

 
__________________________________________________________________________________________________________ 
  Algorithm  4: Recommendation  strategy  based  on  user  incentive- 
  forward rate(BUIRS-G)                                                             _________ 
    Input:   The user uj, worker wi  task rk, candidate pick-up 
           point setPk, road network N 
    Output:   The recommend pick-up point setPk  of task rk 
  1   for each point p ∈ Pk 
  2          Rfp ← Rf ( 
   pprk (wi,uj)) 
 3   end for 
 4   Pk ←∅,sort(Rf),Rf = front(Rf) 
  5   for each point Rfp ∈ Rf 
 6       if p satisfies judgmentconditions, then 
 7          Pk ← Pk ∪{p} 
 8   end for 
 9   return the recommend pick-up point set result Pk    

Simulation environment and data set

The data set used in this article is the data set of Didi Dache in Chengdu, China in November 2016. The experiment uses Python language. The environment of simulation experiment is 64-bit Windows 10 system, memory (RAM) is 16GB, and the processor is Intel(R) Core(TM) i5-9400F CPU @ 2.90 GHz. In order to test the algorithm proposed in this article, part of the data set of Didi Dache in Chengdu, China, on November 1, 2016, was extracted according to the spatial and temporal characteristics of peak hours, including 4,824 order data, including the start and end time of orders and the latitude and longitude of getting on and off the bus. The coordinate system was GCJ-02. The road network data set is the road network vector data of Chengdu, Sichuan Province.

Based on the order information during peak hours, this study intercepts the road network information within the 4th ring road of Chengdu, including 46,661 interchanges and 131,366 road side information. The vector road network information includes road side node information, road grade, road name, one-way and dual-way traffic restriction information and maximum speed limit information. The road network information is shown in Fig. 11.

For the Didi real data set, we use the pick-up and drop-off locations in the order to initialize the location of the user and destination in the pick-up point recommendation question. We process each order in turn according to the start time of processing in the real data set. The waiting time for each order is randomly selected in [4,10]. The location of the driver is randomly generated on the road within a circle with a radius of 1 km from the center of the user; The speed v of each driver is set in the range of [40, maxspeed] according to the road speed limit. Set the maximum movement distance for each user to 400 m.

Performance analysis

Five evaluation indicators are introduced in this section, namely time cost, average response time, energy consumption cost, the number of successful matches and matching success rate.

(1) Analysis of various indicators of pick-up point

This section first shows the selection of a pick-up point for an task in the experiment. The recommended pick-up point of the user u₁ is shown in Fig. 12. The star is the location of the user, the gray is each candidate pick-up point, and the blue is the pick-up point recommended by the algorithm. Figure 13 shows the indicators of the top 12 pick-up points in task execution cost.

As can be seen from Fig. 13, pick-up point p₁ is larger in the walking distance of users, but the difference is not very obvious compared with other candidate pick-up points. Pick-up point p₁’s incentive value is in the middle position in terms of the benefit ratio of walking distance, but it is obviously better than other candidate points pick-up distance. In the fourth indicator of drop-off distance, the pick-up point p₁ also occupies a large advantage, and the algorithm selects the point p₁ in line with expectations.

(2) Analysis of the impact of maximum walking distance on total cost

Figure 14 shows the cost comparison diagram between the user’s original travel distance and the shortest travel distance within each walking distance. Among them, the yellow line segment is the original travel distance, the blue bar is the total travel distance of each optimal pick-up point under different user maximum walking distance restrictions, and the orange line is the actual user walking distance. It can be seen that the total cost of the task is gradually decreasing while the maximum walking distance limit is increasing. This is because when the maximum walking distance of users continues to increase, the number of available pick-up points increases, and users can have better pick-up point selection schemes. In the process of increasing the range from 50 m to 200 m, although the number of candidate pick-up points increases, the new pick-up points cannot further optimize the execution cost, so the choice of pick-up points remains unchanged. The execution cost of the order has not changed either, but as the maximum walking distance of the user has been further increased, better pick-up points have emerged, and the execution cost of the order has decreased.

(3) Analysis of indicators of forward rate

When the maximum walking distance of the user increases, the corresponding number of pick-up points increases exponentially. Figure 15 shows the number of alternative pick-up points of the user under different maximum walking distance restrictions. When the maximum walking distance is increased from 50 to 400 m, the number of candidate pick-up points increases from 0 to 72. When the limit is further expanded, it can be seen that the number of candidate pick-up points increases geometrically.

Because the actual distance calculation in this article does not take the simple Euclidean distance as the distance between two points, but the actual road network distance as the standard. The distance calculation involves the search for the shortest path of the graph and includes the traffic restriction information of each section. The search for the optimal option is not the best choice in calculating the order execution assembly corresponding to each pick-up point under the platform with high real-time requirements, which consumes a lot of computing resources on the platform. In this section, the forward rate index proposed in the second section is experimentally verified.

For the user u₁, under the limitation of the maximum walking distance of 400 m, the relationship between the sequence and the direction rate of each point is obtained by ordering all the candidate pick-up points according to the task distance cost, as shown in Fig. 16. As can be seen from the figure, with the increase of total distance cost, the forward rate of pick-up points decreases. Although a few pick-up points in the middle have higher forward rates, the overall downward trend will not be affected. The objective of this article is to find relatively optimal pick-up points under the condition of optimization of computational load, and if only the optimal pick-up points are selected, the real-time response cannot be satisfied under the condition of a large number of user requests. The candidate pick-up points whose forward rate is lower than a certain threshold can be discarded in subsequent calculation, which greatly reduces the loss of computing resources.

Figure 17 shows the trend chart of task execution cost and forward rate for the first 12 alternative pick-up points of user 1. As can be seen from the figure, the pick-up points with low distance cost have a higher turn-along rate. Therefore, the turn-along rate index proposed in this article can represent the pick-up points’ turn-along rate to a certain extent. When recommending and screening pick-up points to users, the platform can first determine the top several alternative pick-up points according to the pick-up rate, and then continue to calculate other indicators to determine the optimal pick-up points. Save computing resources.

Comparative analysis

In this section, a performance comparison of methods used to solve the pick-up point selection problem is described. Pick-up point recommendation strategy based on user incentive BUIRS-G greedy algorithm and BUIRS-F forward rate algorithm are compared with Song et al. (2017) Three-dimensional matching recommendation strategy (TSMRS) and random recommendation strategy (RRS) in the indicators of task execution cost and running time.

Specifically, in each round of algorithms, the platform selects the best pick-up point for users in each order according to the two recommendation strategies mentioned in this article. In the experiment in this article, BUIRS-G defaults to the pick-up point with the smallest task execution cost as the best pick-up point, and BUIRS-F defaults to the pick-up point with the highest forward rate as the best pick-up point. In contrast, in TSMRS, users, workers and pick-up point are considered to minimize the moving distance between users and workers as the solution goal to obtain the optimal pick-up point. In each round, the algorithm pays attention to the shortest move in the first stage, but ignores the trip information in the second stage and does not consider the overall income of the selected pick-up point, which may lead to high driving cost in the second stage.

Through the selected order data of Didi Travel during peak hours, the influence of the optimal pick-up point selected under different maximum walking distance limits of 0.1 km, 0.2 km, 0.3 km and 0.4 km on the total order cost is set, as shown in the figure.

Figure 18 shows the task optimization ratio of each method under the maximum walking limit of different users. The calculation method is the ratio of the task execution cost under the optimal pick-up point selected by each method to the task execution cost directly through the user’s location. Figure 19 shows the time cost under each method. BUIRS-G proposed in this article is better than the other three methods in terms of cost optimization, followed by BUIRS-F, TSMRS method is not as good as the first two methods in terms of overall optimization because it ignores the task cost in the second stage.

In terms of time cost, BUIRS-G calculates each indicator of candidate pick-up points one by one, so the speed is always slow, which is more obvious when the maximum walking limit increases. Because the number of pick-up points increases significantly, BUIRS-F method only needs to calculate the forward rate, and uses the pick-up points with a higher forward rate to calculate the user incentive value. Therefore, the time overhead is small. TSMRS method needs to calculate the task cost of the first stage one by one, and the running time is smaller than BUIRS-G method. RRS is a random selection of pick-up points, each selection has randomness, and the optimization cost is low. In terms of time overhead, there is no calculation for random selection, so the overhead is the least.