CN117035897B

CN117035897B - Instant delivery platform rider unilateral method

Info

Publication number: CN117035897B
Application number: CN202310620922.1A
Authority: CN
Inventors: 黄鹂强; 季小喻; 季霆; 李玉安
Original assignee: Zhejiang University ZJU
Current assignee: Zhejiang University ZJU
Priority date: 2023-05-26
Filing date: 2023-05-26
Publication date: 2024-04-12
Anticipated expiration: 2043-05-26
Also published as: CN117035897A

Abstract

Aiming at the problem of selecting a delivery order with a time window, the method uses the maximization of the income of a rider as an objective function, and uses the order constraint before and after the delivery point, the time window constraint and the rider load constraint as main constraint conditions to construct a mixed integer planning model; and solving the mixed integer programming model, and taking the optimal solution as a planning scheme of a rider robbing order. The application also provides a GRASP (generic active matrix system) ILS algorithm for solving the mixed integer programming model, which is suitable for a single scene of a crowded-package rider, and has better solving performance and higher convergence speed.

Description

Instant delivery platform rider unilateral method

Technical Field

The application relates to a method for robbing a ticket by a rider, belongs to the field of Internet of things, and particularly relates to a method for robbing a ticket by a rider of an instant distribution platform.

Background

The main function of the instant delivery platform is to effectively match the demand of delivery orders with human resources provided by a rider, thereby generating the benefits of the platform, consumers and the rider. The prior art generally researches order distribution and path optimization problems from the viewpoint of a platform, but relatively few researches from the viewpoint of consumers or riders, and the application mainly aims at the investigation of the robbery of the riders and the distribution planning.

Employment relationships between instant messaging platforms and riders are generally divided into three types, full-time, outsourcing and crowdsourcing modes, respectively. The full-time mode is mainly used for carrying out optimization decision of benefit maximization from the angle of a platform; and because the mode manpower cost is high in this mode, and most instant delivery orders concentrate on the time quantum of having dinner, the fluctuation that the order appears in the rest time is smooth relatively, and the mode that the rider employed full-time entirely is not advisable, can cause the idle of non-peak time quantum rider's capacity resource, causes very big pressure to instant delivery platform enterprise control labor cost. The outsourcing mode is to obtain sufficient capacity by cooperation of an outsourcing mode and enterprises with certain conditions in all areas, and the instant distribution enterprises provide support in aspects of flow, operation, training, excitation and the like. Although the outsourcing mode can rapidly expand service categories at low cost, management of outsourcing also increases costs. The crowdsourcing mode is similar to part-time, and the platform can effectively utilize the capacity resources of the society to be idle to cope with the pressure of the peak period of orders. The payroll of a crowd-sourced rider is typically calculated in a single order, and the price of the order is related to the type of goods and distance that are delivered. Unlike the delivery of full-time and outsourced rider passive acceptance platforms, crowd-sourced riders often need to make their own decisions about whether to pick up a new order when it occurs, how efficiently the order of pick up delivery should be arranged, etc.

Disclosure of Invention

According to one aspect of the application, a method for robbing a single rider by an instant delivery platform is provided, and decisions are made from the viewpoint of benefit optimization of a single rider, so that the actual situation of delivery of the rider is more met. The method and the device ensure the solving quality and greatly improve the solving speed.

The method for robbing the instant delivery platform rider comprises the following steps: taking a timely delivery platform rider order taking as a pick-and-delivery order selection problem with a time window, taking the maximization of the income of the rider as an objective function, and taking the front-and-back order constraint of pick-and-delivery points, the time window constraint and the rider load constraint as main constraint conditions to construct a mixed integer programming model; and solving the mixed integer programming model, and taking the optimal solution as a planning scheme of a rider robbing order. Compared with the existing model, the method has more comprehensive consideration on constraint conditions.

The windowed pick order selection problem is similar to the windowed pick vehicle path problem (Pickup and Delivery Problem with Time Windows, PDPTW), but there is still some difference, mainly as follows:

while the PDPTW considers path planning of multiple vehicles or riders, the objective function is generally global benefit optimization, the problem of picking and delivering orders with a time window is from the perspective of benefit optimization of a single rider, so that only one rider is considered, and in a real situation, the rider can only consider own benefits to make decisions, and other riders are not considered.

2. Unlike PDPTW where all nodes need to be accessed, in the time-windowed pick-and-place order selection problem, the crowd-sourced rider can pick up part of the order to fulfill it completely at his own discretion, i.e., not all nodes need to be accessed.

For the rider, the nodes to be dispatched are divided into two types, one being a pick-up point and one being a delivery point. When a rider accepts an order, the order will appear as a pair of a pick-up point and a delivery point, and the rider must first arrive at the pick-up point to pick up the item and then deliver the picked item to the delivery point. Both the pick-up and delivery points have a time window, and if the rider arrives at the node earlier than the start time of the time window, he must wait until the start of the time window before getting the item, and correspondingly, in a real-time delivery scenario, the rider cannot get the meal if he arrives at the merchant store too early and the merchant store too late to eat it may go bad. In addition, there is an upper limit to the weight or volume that a rider can carry a commodity, so there is some capacity or volume constraint for the rider.

The rider has to start and end the entire service flow at the designated position, the positions of the start point and the end point are not necessarily the same node, but the returning route is not considered by the rider in general, so the delivery point of the order last serviced by the rider is set as the end point.

The purpose of the time windowed pick delivery order selection problem is to make optimal decisions for the crowd-sourced rider to select new orders and route optimization at the same time, in the case that an existing order is being fulfilled. The coordinates of the starting point of the rider, the pick points of all orders, and the delivery points are all known, and the distance between the points is Manhattan. The time window, commodity weight and volume of each order are known and constant, and the running speed and maximum cargo capacity of the rider's vehicle are also known and constant; the commodity does not consider factors such as temperature, deterioration, etc., and uncontrollable factors such as traffic accidents, traffic jams, weather changes, etc. are not considered in the distribution process.

On the instant distribution platform side, a waiting strategy is adopted, namely, orders generated by accumulation are put into APP at the rider end at intervals of t, and the strategy is consistent with the strategy adopted in the actual operation of an enterprise.

In fact, research at home and abroad is mostly not focused on the PDPTW problem itself, but only on how to solve the PDPTW problem.

Preferably, the objective function is established as follows:

max∑ _i∈R p _i ·y _i -∑ _i∈N ∑ _j∈N c _ij ·x _ij

wherein p is _i The service fee for order i; y is _i For the 0-1 variable, when order i is ordered by the rider, y _i =1, otherwise y _i =0; r is the collection of all orders; c _ij Is the distance between node i and node j; x is x _ij As a 0-1 variable, x when the rider travels directly from node i to node j _ij =1, whereas x _ij =0; n is the set of all nodes.

Preferably, the pick-up point front-to-back order constraint comprises:

a rider enters from a node and must come out of the node:

wherein x is _ij As a 0-1 variable, x when the rider travels directly from node i to node j _ij =1, whereas x _ij =0; p is the collection of all the pick-up nodes, D is the collection of all the delivery nodes;

from the starting point, the rider:

∑ _j∈P,j≠0 x _0j =1 wherein 0 represents the starting point of the rider;

the rider returns to the starting point:

wherein 2n+1 represents the rider's end point;

orders that the rider has taken must be fulfilled:

wherein y is _i =1 indicates that order i has been ordered by the rider; r is (r) _r Taking a set of orders for the rider;

orders in the order candidate pool are selected by the rider:

wherein y is _i As 0-1 variable, when the order candidate pool R _s When order i in (a) is ordered by a rider, y _i =1, otherwise y _i ＝0；

If order i is selected by the rider, then the rider must pass the pick point of order i:

if order i is selected by the rider, then the rider must pass the delivery point of order i:

wherein n+i is a delivery node; n is the number of orders; n is the set of all nodes.

Preferably, the time window constraint is as follows:

wherein T is _i Is the time when the rider left node i. t is t _ij The time it takes for the rider to go from node i to node j. n+i is the delivery node, n is the number of orders; x is x _ij For the 0-1 variable, if the rider is traveling directly from node i to node j, then x _ij =1, whereas x _ij =0; n is the set of all nodes. T (T) _ij The method is characterized in that the method is a large M constant and is used for converting time window constraint into equality constraint or inequality constraint in linear programming, meanwhile, the condition that an infeasible solution or an unlimited solution occurs in the solving process can be avoided by using the large M constant, and therefore feasibility and correctness of a planning model are guaranteed. a, a _i Upper time window for node i, b _i Is the lower time window for node i.

Preferably, the rider load constraints include a maximum capacity constraint, as shown in the following equation:

wherein Q is _i The load when the rider leaves the node i; q is the maximum load that the rider can bear; d, d _i Is the load of order i; x is x _ij For the 0-1 variable, if the rider is traveling directly from node i to node j, then x _ij =1, whereas x _ij =0; n is the set of all nodes.

Preferably, a grasp_ils algorithm is adopted when the mixed integer programming model is solved, the grasp_ils algorithm adopts a meta heuristic algorithm framework, and random disturbance is added to a local search operator part of the GRASP algorithm.

The prior art generally adopts a heuristic algorithm (such as a genetic algorithm), but compared with search type algorithms such as ALNS, the algorithm needs a more severe convergence condition, and may not be suitable for processing complex actual data. The ALNS algorithm, although capable of producing a better solution to the PDPTW problem than other algorithms, is not designed to be commercially viable, and therefore, is difficult to obtain a good quality result when applied to the instant distribution crowd-sourced rider robbery scenario of the present application.

Preferably, the initial solution construction of the grasp_ils algorithm includes: firstly, constructing an empty solution which only comprises a starting point and a terminal point of a rider, and then dividing the orders into two groups, wherein one group of orders is an order received by the rider and corresponds to a set of received orders; one group is orders which can be selected by a rider to take, and corresponds to an order candidate pool;

based on this null solution, a deterministic initial solution and a stochastic initial solution are constructed, respectively.

Preferably, the method for constructing the deterministic initial solution includes:

the orders in the set of taken orders are tried to be inserted into the tail end of the existing path of the rider one by one, each order generates an objective function increment, and finally, the order with the largest objective function increment is selected to be inserted and then a new round of circulation is carried out until all the taken orders are inserted into the path of the rider;

orders in the order candidate pool are inserted in the same manner until either the time window constraint or the rider load constraint is violated and insertion cannot continue.

Preferably, the construction method of the random initial solution includes:

the orders in the set of taken orders are tried to be inserted into the random positions of the existing paths of the rider one by one, each order generates an objective function increment, and finally, the order with the largest objective function increment is selected to be inserted and then a new round of circulation is carried out until all the taken orders are inserted into the paths of the rider;

Preferably, the local search operator LS includes at least any one of the following: a single node repositioning operator, an order exchange operator, an order insertion operator and an order removal operator; and finally, selecting one local search operator which enables the objective function to be optimal to operate after the operator is called by the local search algorithm.

Preferably, the single-node relocation operator refers to: moving a single node in the path, moving the node forward or backward in the order in the existing path, but the movement cannot violate any constraint, otherwise the movement is invalid; after the node moves, a certain positive or negative increment is generated on the objective function value.

Preferably, the order exchange operator refers to: the order of a pair of orders is exchanged in the path, and a certain positive or negative increment is generated on the objective function value after the order is exchanged.

Preferably, the order insertion operator refers to: and randomly selecting an order from the order candidate pool, inserting the order into the path of the existing rider, performing traversal attempt on the inserted position, and finally selecting the optimal position for insertion.

Preferably, the order removal operator refers to: an operator that selects an order from the existing path to remove, and the order must be an order that the rider has not received; all the orders which are not received are tried one by one, and finally, the optimal order is selected for removal, wherein the optimal order refers to the maximum improvement value of the objective function after the order is removed.

Preferably, when the local search operator portion of the GRASP algorithm adds random perturbations, a random perturbation operator is employed that sequence swaps two nodes on arbitrarily selected paths, but the sequence swaps must not violate time window constraints, rider load constraints, and order constraints before and after the delivery point is fetched.

The beneficial effects that this application can produce include:

1) The method and the system consider timely delivery platform rider order taking as a pick-and-place order selection problem with a time window, and establish a mathematical model by taking the maximum benefit of the rider as an objective function. Firstly, the model starts from the rider demand side, but not the platform side, fills the gap in the path planning problem of the instant distribution crowdsourcing rider in the existing research, and is more in line with the actual situation.

2) The application aims at the established mathematical model, provides a GRASP ILS algorithm for solving, and the algorithm is specially designed for the industry of instant distribution crowdsourcing riders, so that compared with the existing ALNS algorithm and the like, the method is more fit to the problem, and the solution obtained by the method is more accurate.

3) Compared with an ALNS algorithm, the GRASP ILS algorithm provided by the application is faster. In a more specific scenario, the modified version of the grasp_ils algorithm may have a speed boost of more than 10 times that of the ALNS algorithm. On the premise of reaching the accuracy of the same solution, the ALNS algorithm needs to consume more than one time; while the grasp_ils algorithm can often find better solutions than ALNS and open source solvers under the same computation time constraints.

Drawings

FIG. 1 is a schematic illustration of a crowd-sourced rider planning problem for rider existing orders 1 and 2 in one embodiment of the present application;

FIG. 2 is a schematic illustration of a crowd-sourced rider planning problem with rider selection to accept orders 3 and 4 in one embodiment of the present application;

FIG. 3 is a schematic diagram of single-node relocation operator path movement in an embodiment of the present application;

FIG. 4 is a schematic diagram of order exchange operator path movement in one embodiment of the present application;

FIG. 5 is a schematic diagram of order insertion operator path movement in one embodiment of the present application;

FIG. 6 is a schematic diagram of order removal operator path movement in one embodiment of the present application.

Detailed Description

The present application is described in detail below with reference to examples, but the present application is not limited to these examples.

Crowd-sourced riders often encounter the following described scenarios in everyday work: two orders are ready to be delivered on hand, three orders that can be accepted are presented on the rider side APP, and the decision the rider is faced with is to select which of the three orders to pick up. Some orders may be very close to the order to be dispatched by the rider and even share a pick-up or delivery point with the order to be dispatched, which may be of great interest, and some orders, although paid to the rider for a very high performance service, may be of a very great distance from the existing order, may take a lot of time and effort even after taking the orders, and may lose some of the opportunity to select other orders, which may be of a much greater cost than the performance service of the remote order.

As shown in FIG. 1, in one embodiment, the same numbers represent the same order, and the arrows represent the path or delivery order that the system plans for the crowd-sourced rider, and the rider already has two orders 1 and 2 before making a decision, at which time there are orders 3, 4, 5 in the system waiting for the order to be preempted by the crowd-sourced rider, at which time the system recommends the most reasonable combination of orders and path planning for the orders 1 and 2 on the rider's hand, so that the rider can maximize his own benefits.

As shown in fig. 2, assuming 50 yuan for order 1, 150 yuan for order 2, 100 meters for the overall path, 1 yuan per meter of delivery cost, then 50+150-100=100 yuan for the rider's return, before providing the robbery decision advice. After the system provides the decision advice: the return of order 1 is 50 yuan, the return of order 2 is 150 yuan, the return of order 3 is 100 yuan, the total path is 200 meters, the distribution cost per meter is 1 yuan, and the return of the rider is 50+150+100-150=150 yuan. The rider increases the revenue by 50 yuan after accepting the decision advice of order 3.

The utility model provides a method for the timely delivery platform to rob the orders by the rider, which is characterized in that the timely delivery platform to rob the orders is regarded as a picking and delivering order with a time window to select, the benefit of the maximized rider is taken as an objective function, and the front and back order constraint of the picking and delivering points, the time window constraint and the rider load constraint are taken as main constraint conditions to construct a mixed integer planning model; and solving the mixed integer programming model, and taking the optimal solution as a planning scheme of a rider robbing order.

The mathematical model comprises:

(one) establishing an objective function

The objective function is to maximize the benefit of the rider, and is specifically composed of two parts, wherein the first part is the service fee of the execution of all orders, and the second part is the distance the rider needs to travel for the execution. As shown in formula (1), p _i Service fee for order i, y _i Is a 0-1 variable, y if order i is accepted by the rider _i =1, otherwise y _i ＝0，c _ij Is the distance, x, between node i and node j _ij Is a 0-1 variable, x is the number if the rider is traveling directly from node i to node j _ij =1, whereas x _ij ＝0。

max∑ _i∈R p _i ·y _i -∑ _i∈N ∑ _j∈N c _ij ·x _ij (1)

(II) construction constraints

A rider enters from a node and must come out of the node:

starting from the starting point:

∑ _j∈P,j≠0 x _0j return to start point=1 (3):

∑ _{i∈D,i≠2n+1} x _i,2n+1 ＝1 (4)

orders that the rider has taken must be fulfilled:

orders in the order candidate pool may or may not be selected by the rider:

equations (9) through (11) are time window constraints:

equations (12) through (13) are maximum capacity constraints:

the remaining constraints:

the symbols in the above formulas (1) to (17) are defined as shown in the following table 1.

TABLE 1 symbol definition

And when the mixed integer programming model is solved, a GRASP (generic generalized algorithm) is adopted, a meta heuristic algorithm framework is adopted by the GRASP (generic generalized algorithm) and random disturbance is added to a local search operator part of the GRASP algorithm.

The grasp_ils algorithm consists of a GRASP algorithm and an ILS algorithm. The generic name of the GRASP algorithm is Greedy Randomized Adaptive Search Procedure (greedy random adaptive search algorithm) and the generic name of the ILS is Iterative Local Search (iterative neighborhood search algorithm).

Assuming that a minimization problem is solved, S and f (S) represent the objective function values of a solution and solution, respectively, of the problem, H is a deterministic greedy algorithm, HR is a greedy algorithm with randomness, and Mutate represents a crossover operator with randomness, similar to the mutation operator in genetic algorithms. LS represents the local search operator, S and f=f (S) represent the optimal solution found by the GRASP x ILS meta heuristic framework and the objective function value of its optimal solution, respectively. For the parameters of the above algorithm, there are generally two: np represents the number of different initial solutions used by the algorithm, ni represents the number of times a local search operator is used in each initial solution to explore a better solution, so the number of times a local search operator is invoked in a GRASP x ILS algorithm is ncls=np x ni.

The GRASP algorithm is a simpler meta-heuristic algorithm, and can be regarded as a local search with multiple departure points (multi-start), each initial solution is randomly generated by a random initial solution constructor, then local optimization is performed by a local search operator, and if a solution better than the current optimal solution is found in the searching process, the algorithm records the objective function value of the current optimal solution and the structure of the solution. The number of loops of the GRASP algorithm may be defined as np >1, ni=1.

As shown in table 2 below, is the logic that runs with the GRASP algorithm described in the form of a pseudo code. A flowchart of one-step iteration of the GRASP algorithm may be referred to in fig. 4.

TABLE 2

The ILS algorithm is also a simpler meta-heuristic algorithm, which first needs to construct an initial solution by a deterministic heuristic algorithm, then uses a local search operator to continuously optimize the initial solution, and adds random disturbance in the process of iteration of the local search operator, so that the local search operator cannot continuously perform repeated search in-situ turning. The ILS algorithm may be described as np=1, ni >1.

Random disturbance is a key point of an ILS algorithm, if the random disturbance is too weak, a local search operator cannot jump out of a region of a local optimal solution, and if the random disturbance is too strong, the fields searched by the two local search operators are irrelevant, which is basically equivalent to a GRASP algorithm. Ideally, the random perturbation would have a moderate intensity setting, according to the most recent domain optimization principle introduced by Glover and Laguna.

As shown in table 3 below, is the logic that runs the ILS algorithm described in the form of pseudo code.

TABLE 3 Table 3

Since the GRASP algorithm and the ILS algorithm have natural structural complementarity, the two algorithms can be effectively combined. The ILS algorithm is mainly in the LS part of the GRASP algorithm, and random disturbance is provided for the GRASP algorithm, so that the GRASP algorithm cannot fall into local optimum.

As shown in table 4 below, the logic of the grasp_ils algorithm operation is described in the form of a pseudo code.

TABLE 4 Table 4

The above-described grasp_ils algorithm has the following advantages over the ALNS algorithm that is more efficient on PDPTW:

1. faster, in more specific scenarios, the improved version of the GRASP x ILS algorithm may have a speed boost of more than 10 times compared to the ALNS algorithm. As can be seen from the experimental results in the fifth chapter, the ALNS algorithm takes more than one time to reach the accuracy of the same solution compared to the algorithm proposed in the present application.

2. The quality of the GRASP ILS solution is guaranteed to a certain extent, and under the same calculation time limit, the GRASP ILS can solve better solutions than the ALNS and the open source solver.

In one embodiment, the initial solution construction of the graps_ils algorithm includes:

firstly, constructing an empty solution which only comprises the starting point and the end point of a rider, and then dividing the orders into two groups, wherein one group of orders is the orders received by the rider, and the other group of orders is the orders which can be selected by the rider to receive the orders. Based on this null solution, it will be explained how to construct deterministic initial solutions and stochastic initial solutions, respectively.

The node order and final objective function values of the deterministic initial solution construction initial solution generated each time are both constant. The method comprises the steps of firstly inserting orders with received orders, firstly trying the orders in a received order set to be inserted into the tail end of an existing path of a rider one by one, generating an objective function increment for each order, finally selecting the order with the largest objective function increment, and then performing a new round of circulation after inserting the order with the largest objective function increment until all received orders are inserted into the path of the rider. The same way then begins to insert orders in the rider-selected order set until either the time window constraint or the rider load constraint is violated and insertion cannot continue.

Unlike deterministic initial solution constructs, stochastic initial solution constructs have some randomness. The method comprises the steps of firstly inserting orders with received orders, firstly trying the orders in a received order set to be inserted into random positions of the existing paths of the riders one by one, generating an objective function increment for each order, finally selecting the order with the largest objective function increment, and then performing a new round of circulation after inserting the order with the largest objective function increment until all received orders are inserted into the paths of the riders. The same way then begins to insert orders in the rider-selected order set until either the time window constraint or the rider load constraint is violated and insertion cannot continue.

In one embodiment, the local search operator LS is designed for the rider robbery problem, including but not limited to the following 4 operators. The 4 operators are tried one by one when called by the local search algorithm, and finally the optimal one is selected for operation.

1) Single node relocation operator: a single node will be moved in the path, moving the node forward or backward in the order of the existing path, but the movement cannot violate any constraints, otherwise the movement is invalid. After the node moves, a certain positive or negative increment is generated on the objective function value. As shown in FIG. 3, for a typical repositioning process, assume that there are constraints between order pick up delivery points # 3, and that delivery point # 1 cannot be moved between order pick up delivery points # 3 because of the constraints to be guaranteed.

2) Order exchange operator: a pair of orders are exchanged in the path, and after the order exchange, a certain positive or negative increment is generated for the objective function value. As shown in fig. 4, the order exchange operator exchanges the order of order No. 3 and order No. 1 so that the overall objective function value is changed.

3) Order insertion operator: an order is randomly selected from the order candidate pool and inserted into the path of the existing rider, the inserted position needs to be traversed, and finally the optimal position is selected for insertion. Order number 2 is newly inserted in the process of completing order number 1 as shown in fig. 5.

4) Order removal operator: is an operator that selects an order from the existing path to remove, and the order must be an order that the rider has not yet received. All the orders which are not received are tried one by one, and finally the optimal order is selected for removal. As shown in fig. 6, a group of orders represented by pick-up point 2 and delivery point 2 are removed from the overall path without passing through the nodes of order No. 2 when order No. 1 is completed. The operator can be contrasted with the operator, and the operator are opposite processes.

In one embodiment, the random perturbation operator is employed when the local search operator portion of the GRASP algorithm adds random perturbation.

The random perturbation operator will arbitrarily choose two nodes on the path for order exchange, but the order exchange must not violate the time window constraints, rider load constraints, and order constraints before and after the delivery point. Unlike the local search operator, the algorithm still accepts the exchange of the set of nodes even if the effect on the objective function value is negative after the random perturbation operator performs node exchange, because the purpose of the random perturbation operator is not to improve the solution, but to help the local search operator jump out of the locally optimal solution.

In one embodiment, certain tests and evaluations are made of the optimal performance of the grasp_ils algorithm.

As a benchmarking scheme for algorithmic performance testing, google's OR-tools open source solver and adaptive large scale random neighborhood search (ALNS, adaptive large neighborhood search) will be used as a comparison.

(1) Computer configuration of experiment

Intel (R) Core (TM) i5-8250U CPU@1.60GHz was used.

(2) Example configuration

Performance evaluation and testing was performed using a well-accepted set of examples from the academy of PDPTW problems. While the standard calculation example of PDPTW is proposed by Li and Lim (2001), the standard calculation example set of Li and Lim (2001) cannot be directly applied to the problem of the robbery of the rider in the study, and a certain degree of improvement is required to be applied to the model proposed in the third chapter.

First, all orders in the Li and Lim (2001) standard dataset must be fulfilled, but orders in the rider robbery problem are divided into two parts, the order set for taken and the order set for to be taken, so the orders in the standard dataset need to be randomly divided into the above two groups. The other is that the orders in the standard dataset have only coordinate information, no price information for the orders, and each order has a service performance price in the case of a rider robbery. For sufficient randomness, the price of the order is generated by multiplying the sum of Manhattan distances of the pick and delivery points, respectively, from the rider's origin by a random coefficient of 0 to 1.

(3) Time performance test

Factors affecting the running time of the grasp_ils algorithm are mainly in terms of both problematic scale and iteration number, and therefore this section tests both aspects using the modified example based on the Li and Lim (2001) standard example in 5.1.1. To more fully test the performance of the algorithm, the tests were performed in three major groups, 100, 500 and 1000, respectively, resulting in the time performance test results of table 5 below.

TABLE 5

From Table 5, it can be initially concluded that the initial solution of the GRASP ILS algorithm is generated only in relation to the size of the order, the more orders the more initial solutions are generated the more time consuming. The time for solving the algorithm is related to the order scale and the iteration times, and the larger the order scale is, the more the iteration times are, and the more the algorithm consumes more time.

(4) Optimizing performance testing

And comparing the ALNS algorithm with the GRASP ILS algorithm by taking the optimal solution OR the reference solution obtained by the Google OR-Tools as a reference standard to obtain a comparison result of the optimal performance. The reason for adopting ALNS as the comparison algorithm is that the ALNS is a set of general framework for solving the vehicle scheduling problem, and the ALNS has excellent performance on the vehicle path planning problems such as VRPTW/CVRP/PDPTW and the like. The principle of the ALNS algorithm is that a series of destroying operators and reconstruction operators are provided, the two groups of operators continuously reform the initial solution of the problem, if one group of destroying operators or reconstruction operators have positive effect on the objective function of the solution, the score of the operators is increased, otherwise, the score is unchanged or reduced, and the destroying operators and the reconstruction operators with strong capability can be screened out through multiple iterations in the mode, so that the ALNS algorithm is more and more efficient.

The first set of test cases is 12 cases of lr101-lr112, each case having about 50 orders, about 100 nodes, and about 1 rider for delivery capacity 200.

The second set of test cases is a total of 10 cases LC1_2_1-lc_1_2_10, each case having about 100 orders, about 200 nodes, about 1 rider, and a delivery capacity of 200.

The third set of test cases is a total of 10 cases LC2_2_1-lc_2_2_10, each of which has about 100 orders, about 200 nodes, about 1 rider, and a delivery capacity of 700.

The results of the optimized performance test are obtained as shown in table 6 below.

TABLE 6

The solution obtained by the OR-Tools is taken as a reference solution, three different calculation examples are used for comparing the ALNS algorithm with the GRASP ILS algorithm, and the distances between the solution obtained by the GRASP ILS algorithm and the reference solution are basically closer than those of the ALNS in the same operation time, and particularly, the solution has better optimization capability on the problem of higher solution difficulty of the group of the Lc2_2_1-Lc2_2_10.

In addition, when the same solving precision is achieved, the time required by the GRASP ILS algorithm is obviously better than that of the ALNS algorithm; it can be observed from table 5.2 that the time difference between ALNS algorithm and GRASP x ILS algorithm is not large if there is no explicit requirement for the difference between the numerical solution and the reference solution; however, as the gap is further reduced, the convergence rate of the ALNS algorithm is obviously reduced, and particularly on the lr101-lr112 example, the same gap average value is achieved, and the time required by the ALNS algorithm is 2 times that of the GRASP ILS algorithm described in the application, so that the algorithm described in the application has higher convergence rate and is more suitable for single scene robbing of crowded riders with high timeliness requirements.

(5) Simulation experiment

The foregoing experiments demonstrate the computational temporal performance and optimization performance of the GRASP x ILS algorithm. The effectiveness of the GRASP ILS algorithm in the problem of rider robbery is demonstrated by simulation experiments as follows. The effectiveness of the problem of the order taking of the riders is compared from two aspects, on one hand, the order taking rate of the instant distribution platform is improved, if the riders only take orders with relatively close distances, some orders with relatively far distances are free of order taking performance of the riders, the experience of consumers is damaged, and therefore improvement of the order taking rate is of practical significance to the instant distribution industry. On the other hand, the problem of the robbery of the riders is to promote the income of the riders from the model or from the practical significance, if the income of the riders is not promoted, the original purpose of the study is lost, and the situation of win-win of the income promotion of the riders and the order receiving rate promotion of the platform cannot be formed. Therefore, the platform order receiving rate and the rider income under different order robbing strategies are compared through a series of simulation experiments, and the simulation experiment results are shown in the following table 7.

TABLE 7

The experimental result shows that the order receiving rate of the instant distribution platform and the income of the rider are obviously improved by utilizing the GRASP ILS algorithm to help the rider to rob the order.

The foregoing description is only a few examples of the present application and is not intended to limit the present application in any way, and although the present application is disclosed in the preferred examples, it is not intended to limit the present application, and any person skilled in the art may make some changes or modifications to the disclosed technology without departing from the scope of the technical solution of the present application, and the technical solution is equivalent to the equivalent embodiments.

Claims

1. The utility model provides a method for instant delivery platform rider to rob the bill, which is characterized in that the method comprises the following steps: taking a rider order taking and delivering order of the instant delivery platform as a take-and-deliver order selection problem with a time window, taking the maximization of the income of the rider as an objective function, and taking the constraint of the order, the constraint of the time window and the constraint of the load of the rider before and after the delivery point as main constraint conditions to construct a mixed integer planning model; solving the mixed integer programming model, and taking the optimal solution as a planning scheme of a rider robbery order;

the establishment of the objective function is shown in the following formula:

maxΣ _i∈R p _i ·y _i -∑ _i∈N ∑ _j∈N c _ij ·x _ij

wherein p is _i A fulfillment services fee for the ith order; y is _i For the 0-1 variable, y when the ith order is ordered by the rider _i =1, otherwise y _i =0; r is the collection of all orders; c _ij Is the distance between the ith node and the jth node; x is x _ij As a 0-1 variable, x when the rider travels directly from the ith node to the jth node _ij =1, whereas x _ij =0; n is a set of all nodes;

the pick-and-delivery point fore-and-aft order constraint includes:

a rider enters from a node and must come out of the node:

wherein x is _ij As a 0-1 variable, x when the rider travels directly from the ith node to the jth node _ij =1, whereas x _ij =0; p is the collection of all the pick-up nodes, D is the collection of all the delivery nodes;

from the starting point, the rider:

∑ _j∈P,j≠0 x _0j ＝1

wherein 0 represents the starting point of the rider;

the rider returns to the starting point:

∑ _{i∈D,i≠2n+1} x _i,2n+1 ＝1

wherein 2n+1 represents the rider's end point; n is the number of orders;

orders that the rider has taken must be fulfilled:

wherein y is _i =1 indicates that the ith order has been ordered by the rider; r is R _r Taking a set of orders for the rider;

orders in the order candidate pool are selected by the rider:

wherein y is _i As 0-1 variable, when the order candidate pool R _s When the ith order in (a) is ordered by a rider, y _i =1, otherwise y _i ＝0；

If the ith order is selected by the rider, the rider must pass the pick point of the ith order:

if the ith order is selected by the rider, the rider must pass the delivery point of the ith order:

wherein n+i is a delivery node; n is a set of all nodes;

the time window constraint is as follows:

wherein T is _i Time when the rider left the ith node; t is t _ij Time required for a rider to spend from the ith node to the jth node; n+i is the delivery node, n is the number of orders; x is x _ij For a 0-1 variable, if the rider is traveling directly from the ith node to the jth node, x _ij =1, whereas x _ij =0; n is a set of all nodes; t (T) _ij Is a large M constant; a, a _i Upper time window for ith node, b _i A lower time window for the ith node;

the rider load constraints include maximum capacity constraints as shown in the following equation:

wherein Q is _i The load when the rider leaves the ith node; q is the maximum load that the rider can bear; d, d _i Load for the ith order; x is x _ij For a 0-1 variable, if the rider is traveling directly from the ith node to the jth node, x _ij =1, whereas x _ij =0; n is the set of all nodes.

2. The instant delivery platform rider ordering method of claim 1, wherein a GRASP ILS algorithm is employed when solving the mixed integer programming model, the GRASP ILS algorithm employs a meta heuristic framework, and random perturbation is added to a local search operator portion of the GRASP algorithm.

3. The instant messaging platform rider preemption method of claim 2, wherein the initial solution construction of the GRASP x ILS algorithm comprises: firstly, constructing an empty solution which only comprises a starting point and a terminal point of a rider, and then dividing the orders into two groups, wherein one group of orders is an order received by the rider and corresponds to a set of received orders; one group is orders which can be selected by a rider to take, and corresponds to an order candidate pool;

4. The instant delivery platform rider preemption method of claim 3, wherein the method of constructing the deterministic initial solution comprises:

5. The instant distribution platform rider ordering method according to claim 3, wherein the method for constructing the random initial solution comprises:

6. The method for robbery of the instant distribution platform rider according to claim 2, wherein the local search operator LS at least comprises any one of the following: a single node repositioning operator, an order exchange operator, an order insertion operator and an order removal operator; and finally, selecting one local search operator which enables the objective function to be optimal to operate after the operator is called by the local search algorithm.

7. The instant messaging platform rider preemption method of claim 6, wherein the single node relocation operator is: moving a single node in the path, moving the node forward or backward in the order in the existing path, but the movement cannot violate any constraint, otherwise the movement is invalid; after the node moves, a certain positive or negative increment is generated on the objective function value.

8. The instant messaging platform rider ordering method of claim 6, wherein the order exchange operator is: the order of a pair of orders is exchanged in the path, and a certain positive or negative increment is generated on the objective function value after the order is exchanged.

9. The instant messaging platform rider ordering method of claim 6, wherein the order insertion operator is: and randomly selecting an order from the order candidate pool, inserting the order into the path of the existing rider, performing traversal attempt on the inserted position, and finally selecting the optimal position for insertion.

10. The instant messaging platform rider ordering method of claim 6, wherein the order removal operator is: an operator that selects an order from the existing path to remove, and the order must be an order that the rider has not received; all the orders which are not received are tried one by one, and finally the optimal order is selected for removal.

11. The instant delivery platform rider order picking method of claim 2, wherein when the local search operator portion of the GRASP algorithm adds random perturbations, a random perturbation operator is employed that sequence-swaps two nodes on an arbitrarily selected path, but the sequence-swaps must not violate time window constraints, rider load constraints, and order constraints before and after the delivery point is fetched.