A multi-task optimization method and system for intercity online car-hailing route planning

Abstract

The present invention relates to a multi-task optimization method and system for intercity online car-hailing path planning, which defines the intercity online car-hailing path planning problem as a multi-objective problem including five objectives; constructs an initial non-dominated planning scheme for the current path planning task through a series of path construction methods based on greed; the multi-task environment construction method based on the problem decomposition strategy decomposes a single path planning problem into multiple sub-problems representing different directions, and combines multiple sub-problems into multiple tasks to achieve multi-tasking, accelerating the process of generating a path planning scheme; and proposes a knowledge transfer strategy to transfer the excellent path planning scheme solutions obtained from the solution regions represented by other different path planning tasks to the target task, and optimize the diversity of the path planning scheme. Finally, an adaptive local search based on a chain is introduced to iteratively optimize the path planning scheme. The present invention can efficiently plan the travel path of intercity online car-hailing.

Description

Technical Field

[0001] The present invention relates to the two technical fields of intelligent optimization calculation and scheduling planning, and particularly relates to a multi-task optimization method and system for intercity online car-hailing path planning. Background Art

[0002] With the rapid development of the economy, the demand for medium and short-distance intercity tourism has been increasing continuously. Travelers usually drive by themselves or take intercity public transportation (such as long-distance buses and railways). In recent years, intercity online car-hailing travel, as a new type of fast transportation mode, has developed in China and is increasingly accepted by the public. The operation mode of intercity online car-hailing is an innovative application under the growing transportation demand of citizens, responding to the call of the country's "Internet +" convenient transportation to promote the development of intelligent transportation. In this travel mode, several passengers from the same city take a car to another city. Compared with transitional public transportation, intercity online car-hailing travel can provide flexible and time-saving door-to-door services, better meeting the personalized travel needs of passengers.

[0003] Different from common car-hailing services, intercity online car-hailing services are mainly to solve the problem of personalized travel between specific cities. Different from in-city travel, intercity travel has the following characteristics: low travel demand, low travel frequency, long travel distance, and few travel options. These characteristics make intercity travel usually highly planned. Therefore, in intercity online car-hailing services, the path planning problem refers to the optimization problem of planning reasonable pick-up and drop-off paths for a series of passengers departing from a certain city to the destination city under the condition of meeting certain constraint conditions and the travel needs of customers between specific cities, so as to reduce the empty-haul rate of vehicles, shorten the waiting time of customers, and reduce the detour distance of drivers, etc.

[0004] The intercity online car-hailing path planning problem belongs to a variant of the Vehicle Routing Problem (VRP). The vehicle routing problem is a classical combinatorial optimization problem and has broad application prospects in many fields such as logistics transportation, supply chain management, production scheduling, and autonomous driving. Currently, many effective solutions to the vehicle routing problem have been proposed in the academic community. However, there is a main problem in the existing research on multi-objective vehicle routing problems: ignoring the information sharing and communication between different problems and sub-problems. Therefore, for most existing methods, effectively solving the optimization problem of multiple conflicting objectives is still a huge challenge.

[0005] The problem of travel route planning for intercity online car-hailing is similar to the vehicle routing problem and has the problem attributes of strong constraints with five objectives. At present, most of the work on travel route planning for intercity online car-hailing is based on single tasks, and it is inefficient to solve the vehicle routing problem with route balance through single-task optimization methods, ignoring the similarity characteristics of the problem. Therefore, how to provide an efficient route planning method and system for intercity online car-hailing services has very important practical significance and application prospects. Summary of the Invention

[0006] The object of the present invention is to provide a multi-task optimization method and system for intercity online car-hailing route planning, which can not only efficiently plan the travel routes of intercity online car-hailing, but also use the multi-task multi-objective optimization method to provide a set of high-quality planning solutions for different needs of intercity online car-hailing services.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A multi-task optimization method for intercity online car-hailing route planning, including:

[0009] S1. According to the actual needs of intercity online car-hailing travel route planning, model the travel route planning problem; the established model is a multi-objective optimization problem with five objectives min f = {f1, f2, f3, f4, f5}, where f1 is the number of online car-hailing vehicles, f2 is the total driving distance of online car-hailing vehicles to pick up and drop off passengers; f3 is the driving time of the longest driving route for online car-hailing vehicles to pick up and drop off passengers, f4 is the total waiting time of online car-hailing vehicles due to arriving at the pick-up and drop-off locations in advance, and f5 is the total delay time caused by online car-hailing vehicles arriving late.

[0010] S2. Judge whether there is an unprocessed online car-hailing order currently. If there is, go to step S3; otherwise, enter the waiting state until a new order arrives and enter step S3;

[0011] S3. Generate an initial planning solution for the order travel route through a greedy path construction method, and save the initial planning solution to the external archive solution set POP.

[0012] S4. According to the initial planning solution, decompose the multi-objective optimization problem into N single-objective sub-problems by using a problem decomposition strategy, and combine the N sub-problems into K tasks;

[0013] S5. Set the k-th task among the K tasks as the target task T k, where \(k = 1, 2, \ldots, K\), and the remaining tasks are defined as auxiliary tasks. The knowledge transfer method is used to optimize among the tasks to obtain a new path planning scheme. According to the new path planning scheme, the first update of the POP is performed using the \(\epsilon\)-dominance archive update strategy to obtain the first POP;

[0014] S6. Perform a second update on all the planning schemes in the first POP through chain-based adaptive local search to obtain the second POP;

[0015] S7. According to the needs of the passengers in the order, select a path for the passengers from the path planning schemes of the second POP and allocate the corresponding vehicles and drivers, and delete the allocated paths from the second POP;

[0016] S8. Set the POP set to an empty set and return to step S2.

[0017] The present invention also provides a multi-task optimization system for intercity online car-hailing path planning, including:

[0018] A model establishment module for modeling the travel path planning problem according to the actual needs of intercity online car-hailing travel path planning; the established model is a multi-objective optimization problem including five objectives \(\min f=\{f_1,f_2,f_3,f_4,f_5\}\), where \(f_1\) is the number of online car-hailing vehicles, \(f_2\) is the total driving distance of online car-hailing vehicles to pick up and drop off passengers; \(f_3\) is the driving time of the longest driving path for online car-hailing vehicles to pick up and drop off passengers, \(f_4\) is the total waiting time of online car-hailing vehicles due to arriving at the pick-up and drop-off locations in advance, and \(f_5\) is the total delay time caused by online car-hailing vehicles arriving late;

[0019] An order judgment module for judging whether there is an unprocessed online car-hailing order currently. If there is, enter the step "Generate an initial planning scheme for the order travel path through a greedy path construction method"; otherwise, enter the waiting state until a new order arrives and enter the step "Generate an initial planning scheme for the order travel path through a greedy path construction method";

[0020] An initial planning module for generating an initial planning scheme for the order travel path through a greedy path construction method and saving the initial planning scheme to the external archive solution set POP;

[0021] A problem decomposition module for decomposing the multi-objective optimization problem into \(N\) single-objective sub-problems according to the initial planning scheme and combining the \(N\) sub-problems into \(K\) tasks;

[0022] A first update module for designating the \(k\)th task among the \(K\) tasks as the target task \(T\) k, where \(k = 1, 2, \ldots, K\), and the remaining tasks are defined as auxiliary tasks. The knowledge transfer method is used to optimize among the tasks to obtain a new path planning scheme. According to the new path planning scheme, the first update of the POP is performed using the \(\epsilon\)-dominant archive update strategy to obtain the first POP;

[0023] A second update module, configured to perform a second update on all the planning schemes in the first POP through chain-based adaptive local search to obtain a second POP;

[0024] A scheme allocation module, configured to allocate each path planning scheme in the second POP to the vehicle and driver corresponding to the path, and delete the allocated path from the second POP;

[0025] An empty setting module, configured to set the POP set to an empty set and return to the step of "judging whether there is an unprocessed online car-hailing order currently".

[0026] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0027] The present invention provides a multi-task optimization method and system for intercity online car-hailing path planning, which defines the intercity online car-hailing path planning problem as a multi-objective problem including five objectives, more comprehensively and realistically reflecting the essence of the path planning problem; constructs an initial non-dominated planning scheme for the current path planning task through a series of path construction methods based on greed; then, the multi-task environment construction method based on the problem decomposition strategy decomposes a single path planning problem into multiple sub-problems representing different directions, and combines multiple sub-problems into multiple tasks to achieve multi-tasking, accelerating the process of generating path planning schemes; at the same time, in order to further utilize the synergistic effects existing between different solution regions, a knowledge transfer strategy is proposed to transfer the excellent path planning scheme solutions obtained from the solution regions represented by other different path planning tasks to the target task, and optimize the diversity of the path planning schemes. Finally, chain-based adaptive local search is introduced to iteratively optimize the path planning schemes. The effective combination of these mechanisms can not only efficiently plan the travel paths of intercity online car-hailing, but also provide a set of high-quality planning schemes that meet different requirements for intercity online car-hailing services using the multi-task multi-objective optimization method. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 Flow chart of the multi-task optimization method for intercity online car-hailing path planning provided in Embodiment 1 of the present invention;

[0030] Figure 2 Example diagram of the operator based on order crossover provided in Embodiment 1 of the present invention;

[0031] Figure 3 Example diagram of the swap operator provided in Embodiment 1 of the present invention;

[0032] Figure 4 Example diagram of the reversal operator provided in Embodiment 1 of the present invention;

[0033] Figure 5 Flow chart of the optimization method provided in Embodiment 1 of the present invention;

[0034] Figure 6 Schematic diagram of the knowledge transfer process provided in Embodiment 1 of the present invention. Detailed implementation manners

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0036] The intercity online car-hailing service involved in the present invention is mainly targeted at the highway passenger transport market, combining the current Internet technology with the car rental industry, making up for the deficiencies of transitional public transportation in personalized travel, and playing an active role in optimizing the allocation of market resources and meeting the high-quality, personalized, and diversified travel needs of citizens. Therefore, efficiently and intelligently planning the travel path of intercity online car-hailing can not only perform vehicle scheduling in advance, bringing a comfortable travel experience to travel customers, but also saving transportation costs and bringing more benefits to the company and drivers. In the intercity online car-hailing industry, the planning of travel path plans is one of the key issues in online car-hailing services.

[0037] The purpose of the present invention is to provide a multi-task optimization method and system for intercity online car-hailing path planning, which can not only efficiently plan the travel path of intercity online car-hailing, but also use the multi-task multi-objective optimization method to provide a set of high-quality planning solutions for different needs of intercity online car-hailing services.

[0038] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0039] Embodiment 1

[0040] This embodiment provides a multi-task optimization method for intercity online car-hailing path planning. Please refer to Figure 1 , the method includes:

[0041] S1. According to the actual requirements of intercity online car-hailing path planning, model the path planning problem; the established model is a multi-objective optimization problem min f = {f1, f2, f3, f4, f5} with five objectives, where f1 is the number of online car-hailing vehicles, f2 is the total driving distance of online car-hailing vehicles to pick up and drop off passengers; f3 is the driving time of the longest driving path for online car-hailing vehicles to pick up and drop off passengers, f4 is the total waiting time of online car-hailing vehicles due to arriving at the pick-up and drop-off locations in advance, and f5 is the total delay time caused by online car-hailing vehicles arriving late.

[0042] In this embodiment, a path planning scheme X is represented by a set O = {o1,..., ok} composed of p routes, where is a path composed of an access sequence including Ni orders and 2Ni pick-up and drop-off points, represents the j-th customer point on the i-th path. Since each order includes the pick-up point of the customer in the departure city and the drop-off point in the target city, therefore, in each path, each order is represented by two pick-up and drop-off points, that is, the pick-up point of the order in the departure city and the drop-off point in the target city. In a path planning scheme, the two pick-up and drop-off points of any order will only appear in one and the same path.

[0043] In the multi-objective model of the intercity online car-hailing path planning problem, it includes 5 different optimization objective functions minf = {f1, f2, f3, f4, f5}, and the specific definitions are as follows:

[0044] f1 = |R| = M

[0045]

[0046] f3 = maX{T p |P = 1,...M}

[0047]

[0048]

[0049] Among them, f1 is used to calculate the number of intercity online car-hailing vehicles, M represents the number of intercity online car-hailing vehicles, and |R| represents the set of paths in the solution. f2 is used to calculate the total driving distance of intercity online car-hailing drivers to pick up and drop off passengers, and Dist p represents the driving distance of intercity online car-hailing vehicle p. f3 is used to calculate the driving time of the longest driving path for intercity online car-hailing vehicles to pick up and drop off passengers, and T pIndicates the driving time of the intercity online car-hailing service p. f4 is used to calculate the total waiting time of the intercity online car-hailing service for arriving at the pick-up and drop-off locations in advance, W p Indicates the waiting time of the intercity online car-hailing service p for arriving in advance. f5 is used to calculate the total delay time caused by the intercity online car-hailing service arriving late, Delay p Indicates the delay time caused by the intercity online car-hailing service p arriving late;

[0050] The five objective definitions described above involve Dist p 、T p 、W p and Delay p The calculations of are as follows. A detailed description is given below.

[0051] 1) Dist p Indicates the total driving distance of the p-th intercity online car-hailing service for picking up and dropping off passengers. The calculation process is as follows:

[0052]

[0053] Among them, and Indicate the i-th pick-up and drop-off point and the (i + 1)-th pick-up and drop-off point of the intercity online car-hailing service p for its trip, Indicates the driving distance of the intercity online car-hailing service p from the i-th pick-up and drop-off point to the (i + 1)-th pick-up and drop-off point of its trip. N p Indicates the set of all intercity online car-hailing services. Since the path planning of intercity online car-hailing services is carried out before the start of the car-hailing service, the actual driving time is limited by the traffic conditions at that time. Therefore, the total driving distance of the intercity online car-hailing service driver for picking up and dropping off passengers in the path planning scheme is calculated by using real-time traffic data when the path planning algorithm is executed.

[0054] 2) T p Indicates the driving time of the p-th intercity online car-hailing service. The calculation process is as follows:

[0055]

[0056] Among them, Indicates the driving time of the intercity online car-hailing service p from the i-th pick-up and drop-off point to the (i + 1)-th pick-up and drop-off point of its trip, Indicates the waiting time of the intercity online car-hailing service p at the pick-up and drop-off point and Indicates the service time of the intercity online car-hailing service p at the pick-up and drop-off point Since the path planning of intercity online car-hailing services is carried out before the start of the car-hailing service, the actual driving time is limited by the traffic conditions at that time. Therefore, the longest driving path for the intercity online car-hailing service to pick up and drop off passengers in the path planning scheme is calculated by using real-time traffic data when the path planning algorithm is executed.

[0057] 3) W p Indicates the waiting time of the intercity online car-hailing p due to arriving early. The calculation process is as follows:

[0058]

[0059] Among them, each customer has a demand q i and a service window [b i , e i . If the vehicle arrives before the earliest service time b i , it must wait until the earliest service time b i to provide service.

[0060] Indicates the earliest service time of the intercity online car-hailing p at the pick-up and drop-off point , and indicates the time when the intercity online car-hailing p arrives at the pick-up and drop-off point . Since the path planning of the intercity online car-hailing is carried out before the car-hailing starts to provide service, the actual driving time is limited by the traffic conditions at that time. Therefore, the total waiting time of the path planning scheme is calculated by using real-time traffic data when the path planning algorithm is executed.

[0061] 4) Delay p Indicates the delay time caused by the late arrival of the intercity online car-hailing p. The calculation process is as follows:

[0062]

[0063] Among them, each customer has a demand q i and a service window [b i , e i . If the vehicle arrives after the latest time e i within the maximum allowable delay time md, it will cause a delay time when arriving at the customer service location.

[0064] Indicates the time when the intercity online car-hailing p arrives at the pick-up and drop-off point , and indicates the latest service time of the intercity online car-hailing p at the pick-up and drop-off point . Since the path planning of the intercity online car-hailing is carried out before the car-hailing starts to provide service, the actual driving time is limited by the traffic conditions at that time. Therefore, the total delay time of the path planning scheme is calculated by using real-time traffic data when the path planning algorithm is executed.

[0065] ​​In addition, the path planning problem of intercity online car-hailing can be described as follows: A group of passengers (denoted as C = {1, 2, …, n}) are served by companies with the same vehicle seat capacity. Each customer has a number of passengers to be carried, i.e., demand q i and a departure time window [b i , e i , and can only be served by one vehicle. Each vehicle belongs to the city where the starting point and the destination are located. The vehicle plans the pick-up and drop-off routes according to the pick-up and drop-off points of each passenger, and sequentially sends each passenger to the corresponding drop-off point. Create a 0-1 decision variable which is equal to 1 if vehicle p travels through path (i, j), and 0 otherwise. And it is required that before any vehicle p picks up the next new passenger, all passengers of each trip need to get off. The present invention adopts a soft time window, that is, within the maximum allowable delay time md, arriving at the service customer location after the latest time e i will result in a delay time dt i . The delay time of the p-th vehicle at the j-th customer point is defined as On the contrary, if the vehicle arrives before the earliest service time b i , it must wait until the earliest service time b i to provide service, which will result in an increase in the waiting time. The waiting time of the p-th vehicle at the j-th customer point is defined as For each vehicle, the total demand of the customers it serves cannot exceed the available seats at this time, i.e., the maximum capacity Q at this time, and it departs and returns to the garage within the service time window [0, e0] of the garage. In addition, considering that the driving time of a single trip for intercity travel is longer than that for intracity travel, it is required that the driver needs to rest between two consecutive trips. In the model of the intercity online car-hailing path planning problem, there are multiple constraint conditions, which are specifically defined as follows:

[0066] 1) Capacity constraint

[0067] The number of passengers in each order cannot exceed the maximum passenger capacity Q of the vehicle, and the total number of passengers of all orders assigned on the same path cannot exceed the maximum passenger capacity Q of the vehicle, that is, it satisfies:

[0068]

[0069] Among them, represents the number of customers of the p-th order of the vehicle on the i-th path .

[0070] 2) Service constraint one

[0071] To ensure the service quality of the online car-hailing, before any vehicle p picks up the next batch of new passengers, it needs to let all passengers of this path get off. That is, it satisfies:

[0072]

[0073] wherein, is a 0-1 decision variable; if vehicle k travels through the arc (i, j), then otherwise

[0074] 3) Service Constraint II

[0075] In a path planning scheme, the pick-up and drop-off points of any order will only appear in one and the same path, and the pick-up point of an order in the departure city must appear before the drop-off point of the order in the destination city.

[0076] 4) Time Constraint

[0077] To ensure the service quality of intercity online car-hailing, the departure time of the vehicle at the agreed pick-up and drop-off points due to traffic, weather or personal reasons shall not exceed the maximum allowable delay time md (in the actual application of the present invention, md can be set according to the intercity online car-hailing service strategy or travel demand, such as set to 30, 45 minutes), that is, it satisfies:

[0078]

[0079] wherein, dt i represents the delay time caused by arriving at the service customer location after the latest time e i represents the delay time of the p-th vehicle at the j-th pick-up and drop-off point.

[0080] 5) Safety Constraint

[0081] Considering that the driving time of a single trip for intercity travel is longer than that for intracity travel, in order to ensure the safety of passengers and drivers during intercity online car-hailing trips and to ensure the travel efficiency of drivers, it is required that drivers take a break between two consecutive trips. That is, it satisfies:

[0082]

[0083] wherein, represents the time when vehicle p starts to serve pick-up point i, s i represents the service time of pick-up point i, t i,j represents the driving time of the vehicle from pick-up point i to pick-up point j.

[0084] S2. Determine whether there is an unprocessed online car-hailing order currently. If so, go to step S3; otherwise, enter the waiting state until a new order arrives, and then go to step S3. ​

[0085] S3. Generate an initial planning scheme for the order travel path through a greedy path construction method, and save the initial planning scheme to the external archive solution set POP.

[0086] In the solution of this embodiment, a greedy sequence path construction method is proposed to generate an initial feasible path planning scheme. In this method, a greedy strategy for serving passengers is proposed to construct feasible vehicle paths in sequence. Since the vehicle capacity for intercity online car-hailing trips is very low, the feasible trips with the maximum profit can be generated by the enumeration method. According to the destinations of all orders in this path and the starting point of the last order in this path, the pick-up and drop-off points are inserted in the order of the distance from near to far. The specific steps are as follows:

[0087] S31. Set the number of path planning schemes i' = 1;

[0088] S32. Obtain all current orders, combine them into an order sequence, obtain the order attributes of all current orders, and obtain a path planning scheme P i′ ; The order attributes include the number of people in the order, the starting point, the starting service time, and the ending point of the order;

[0089] S33. Randomly select an order from the order sequence, denoted as the first order, create a path t according to the starting point and ending point of the first order, and use the starting point of the first order as the starting point of the online car-hailing vehicle p; the online car-hailing vehicle p is the one whose service time (the working time of the driver of vehicle p) meets the starting service time of the first order and is the closest to the starting point of the first order;

[0090] S34. Judge whether the number of people in the first order is equal to the number of empty seats in the online car-hailing vehicle p;

[0091] If so, go to step S310; if not, go to step S35;

[0092] S35. Judge whether there are still unselected orders in the order sequence. If so, go to step S36; otherwise, go to step S310;

[0093] S37. Randomly select an unselected order from the order sequence, denoted as the second order;

[0094] S38. Judge whether the starting service time of the second order meets the service time of the online car-hailing vehicle p to obtain a first judgment result, and judge whether adding the second order to the orders of the online car-hailing vehicle p meets five constraint conditions to obtain a second judgment result;

[0095] If both the first judgment result and the second judgment result are yes, add the second order to the orders of the online car-hailing p; otherwise, return to step S35;

[0096] S39. Determine whether the number of passengers in the orders currently added to the online car-hailing p is equal to the number of empty seats in the online car-hailing p;

[0097] If so, go to step S310; if not, return to step S35;

[0098] S310. Use the greedy strategy to sort the destinations of all orders of the online car-hailing p and the starting points of the remaining orders except the first order, and insert the starting points and destinations of the remaining orders of the online car-hailing p into the path t according to the sorting result to obtain the route planning route of the online car-hailing p, and add the route planning route to the route planning scheme P i′ and set the status of all orders of the online car-hailing p to "processed", and remove them from the order sequence;

[0099] S311. Add the route planning scheme P i′ to the initial route planning solution set POP;

[0100] S312. Determine whether there are unprocessed orders in the order sequence. If so, return to step S33; otherwise, go to step S313;

[0101] S313. Determine whether the number of route planning schemes in POP reaches N. If so, end; if not, execute i' = i' + 1 for i' and return to step S32.

[0102] It should be noted here that it is necessary to make the number of route planning schemes in POP correspond to the number of subsequent N sub-problems, so that the route planning schemes can be optimized later for different sub-problems. Therefore, when the number of route planning schemes in POP does not reach N, it is necessary to return to step S32 to plan a new route scheme. Although the processed orders have been deleted from the order sequence in the aforementioned step S310, when returning to step S32 in step S313, it is considered that all current orders still exist in step S32.

[0103] The following steps S4 - S6 in this embodiment are the optimization methods adopted in this solution, including a multi-task environment construction method based on a problem decomposition strategy, a knowledge transfer method, a chain-based adaptive local search method, and an archive update strategy. The detailed optimization process is introduced below, and reference can be made to Figure 5 for understanding.

[0104] S4. According to the initial planning scheme, the multi-objective optimization problem is decomposed into N single-objective sub-problems by using a problem decomposition strategy, and the N sub-problems are combined into K tasks.

[0105] In this embodiment, a multi-task environment construction scheme based on a problem decomposition strategy is used to decompose a single problem into multiple sub-problems representing different directions, and the multiple sub-problems are combined into multiple tasks to achieve multi-tasking, accelerating the process of generating a path planning scheme.

[0106] Specifically for the problem decomposition strategy, for the intercity online car-hailing travel path planning problem with m (m = 5) objectives, the decomposed sub-problems are defined by reference points. Using the reference boundary intersection method, the m-dimensional solution space Ω is sampled to obtain weight vectors λ = λ1,..., λ N that are evenly distributed in the space, and where H is the number of samples in each objective direction, and its sampling step is δ = 1 / H. The problem decomposition strategy decomposes the 5-objective optimization problem into N sub-problems, and the definition formula 1 of the i-th sub-problem is as follows:

[0107]

[0108] where minf v and maxf v represent the maximum value and the minimum value on the v-th objective respectively. The part represents normalizing the objective v.

[0109] Specifically for the multi-task environment construction, since the intercity online car-hailing path planning problem is a 5-objective optimization problem, for this expensive and complex multi-objective optimization problem, optimizing the multiple decomposed sub-problems is very resource-consuming in terms of search. Through the process of multi-tasking, the N sub-problems are evenly divided into K tasks, and each task includes many adjacent sub-problems. In each task, the fitness value of the central sub-problem is regarded as the representative fitness value of the task. Because the reference point directions of all the component sub-problems are close to each other, the sub-problems in the same task have similarity, and the optimization of the sub-problems is transformed into the optimization of the task. Since these sub-problems are derived from the decomposition of this path planning problem, some tasks can transfer the beneficial knowledge found to other tasks, and then realize the knowledge exchange and sharing between the solution regions of the Pareto front, accelerating the search process of generating a path planning scheme and enhancing the diversity and frontiers of the obtained path planning scheme.

[0110] For the multi-objective model and specific scenarios of the intercity online car-hailing path planning problem, the parameter settings are as follows: In the multi-task scenario based on the problem decomposition strategy, the original problem is decomposed into 70 sub-problems (N = 70), and the number of tasks (K) is set to 5. That is to say, 1 original problem becomes a 5-task form with 14 sub-problems, and each sub-problem has a corresponding weight vector solution. The method for constructing a multi-task environment based on the problem decomposition strategy adopted in this embodiment is specifically as follows:

[0111] S41. Use the boundary crossing method to decompose the multi-objective optimization problem into N sub-problems, form a sub-problem set with the N sub-problems, and initialize N solutions according to the initial path planning scheme to obtain an initial planning solution set; the N solutions correspond one by one to the N sub-problems;

[0112] S42. Randomly select an unselected sub-problem from the N sub-problems, denoted as the first sub-problem, and select the solution with the smallest fitness value in the initial planning solution set corresponding to the first sub-problem as the weight vector solution of the first sub-problem, remove the weight vector solution from the initial planning solution set, and return to the step "Randomly select an unselected sub-problem from the N sub-problems" until the initial planning solution set is empty.

[0113] Step S42 is mainly to make the sub-problems correspond one by one to the initial planning scheme.

[0114] S43. Initialize the number of multi-tasks K, and evenly divide the N sub-problems into K tasks; n = N / K represents the number of sub-problems in each task after the N sub-problems are assigned to K tasks.

[0115] S44. Calculate the center point x of each task, and determine the central sub-problem according to the center point, add the central sub-problem to the corresponding task, and remove the central sub-problem from the sub-problem set;

[0116] Among them, the center point x is calculated by the formula Calculate and obtain.

[0117] S45. Use the Euclidean distance to calculate the n - 1 points closest to each center point x, and determine the sub-problems corresponding to the n - 1 points respectively, denoted as the closest sub-problems, add the closest sub-problems to the task corresponding to the center point x, and remove the closest sub-problems from the sub-problem set.

[0118] S5. Set the kth task among the K tasks as the target task T k, where k = 1, 2,..., K, and the remaining tasks are defined as auxiliary tasks. The knowledge transfer method is used to optimize among the tasks to obtain a new path planning scheme. According to the new path planning scheme, the ε-dominance archive update strategy is used to perform the first update on the POP to obtain the first POP.

[0119] In this embodiment, the knowledge transfer method is used to optimize the diversity of the path planning scheme. Among them, the encoding and population initialization process. Specifically, in this embodiment, integer encoding is used according to the actual situation of intercity online car-hailing travel services to encode pick-up and drop-off points. One chromosome represents a pick-up and drop-off path; the number of genes of one chromosome represents the types of pick-up and drop-off points in the optimization area, and each gene of the chromosome represents a pick-up and drop-off point. Initializing the population is the initial search point. Before optimizing the planned path, the pick-up and drop-off points are classified and sorted, and the pick-up and drop-off points are numbered according to integer encoding. The evolutionary operations involved in the knowledge transfer method are defined by two basic operators: the genetic operator and the environmental selection operator. The genetic operator in this embodiment includes three types (Order_based crossover operator, swap operator, and reverse operator), and the genetic operator defines the best position to insert pick-up and drop-off points in the path. The environmental selection operator is used to update a better path planning scheme. The specific definitions are as follows:

[0120] Order_based crossover operator: (For the example of the order-based crossover operator shown specifically in Figure 2 ) Consider two paths, and randomly select several pick-up and drop-off points in path 1. The positions can be discontinuous. Then find the positions of the pick-up and drop-off points of path 1 selected in path 2, and then generate offspring with the remaining pick-up and drop-off points in path 2. Finally, place the selected pick-up and drop-off points in path 1 in order in the remaining positions of the offspring.

[0121] Swap operator: Randomly swap two pick-up and drop-off points in a path sequence (for the example of the swap operator shown specifically in Figure 3 ).

[0122] Reverse operator: Randomly select a segment of pick-up and drop-off point fragments in a path and reverse the pick-up and drop-off point fragments. As shown in Figure 4 , randomly select a segment of customer point fragments from 2 to 5. After reversing the sequence of {2, 3, 4, 5}, it becomes {5, 4, 3, 2}. The final customer sequence of the obtained individual is Figure 4 shown in the second row of

[0123] Environmental selection operator: Update the weight vector solutions for all sub-problems of the task. In the combined population of parents and offspring, according to formula (1), reselect the solution with the minimum fitness value for each sub-problem of the target task as the weight vector solution for that sub-problem.

[0124] While new solutions emerge, update the external archive POP using the ε-dominance archive update strategy (the specific method will be received in the subsequent plan).

[0125] The three genetic operators exchange chromosomes for all individuals in the population with a set probability value and mutate with a set probability. The environmental selection operator ensures that during the path optimization process, the more optimized paths are retained, while the inferior paths that do not meet the requirements or the paths that do not satisfy the constraints are eliminated. Through gradual inheritance and knowledge transfer, information exchange and sharing between populations are achieved to optimize the adaptability of the next generation of new individuals, making the new approximate solutions better meet the indicators of the intercity online car-hailing travel path planning. Through the above definitions, the specific steps of the knowledge transfer method in this embodiment are as follows (please refer to Figure 6 ):

[0126] S51. Encoding and population initialization;

[0127] Before optimizing the planned path, classify and sort the pick-up and drop-off points, and number the pick-up and drop-off points according to integer encoding. Each chromosome is regarded as a travel path, each individual is regarded as a path planning scheme, and all path planning schemes are regarded as a population C.

[0128] S52. Set the population C to be empty, and add the weight vector solution of the target task T k to the population C, where k = 1;

[0129] S53. Use the NSGA-II algorithm to perform the fast non-dominated sorting of the classical algorithm NSGA-II on the weight vector solutions of each task in the auxiliary task T j , j ∈ {1,…, K}\{k}, and obtain the first front solution Front according to the sorting result;

[0130] S54. Combine the weight vector solution of the target task and the first front solution to obtain a combined path planning scheme, and use the combined path planning scheme as the parent population;

[0131] S55. Determine whether the number of the first front solutions is greater than N / K; if so, go to step S56; if not, add the first front solutions to the population C;

[0132] S56. Randomly select N / K solutions from the first front solutions and add them to the population C;

[0133] S57. Randomly apply genetic operators to the population C to obtain the offspring population C * , and based on the offspring population C * , perform the first update on the POP using the dominant archive update strategy;

[0134] Among them, the genetic operators described in step S57 are the Order_based crossover operator, swap operator, and reverse operator mentioned above.

[0135] S58. Combine the parent population C and the offspring population C * to obtain the combined population C 联 , and delete the duplicate path planning schemes in the combined population C 联 ; where C 联 = C ∪ C * .

[0136] S59. Use the environmental selection operation to reselect the solution with the smallest fitness value as the weight vector solution for the sub-problems corresponding to the target task in the combined population C 联 ;

[0137] S510. Judge whether k is equal to K. If not, perform the operation of k = k + 1 on k and return to step S53; if so, end.

[0138] Among them, the environmental selection operation is to sequentially check the weight vector solutions of all sub-problems of the current target task and the weight vector solutions of all sub-problems of the auxiliary task. In the combined population of the parent and offspring, according to formula (1), reselect the solution with the smallest fitness value as the weight vector solution for each sub-problem of the target task. of the weight vector solution is updated by C index , remove the solution C from C index , and subtract 1 from the size of the population C.

[0139] Among them, the Order_based crossover operator means considering two paths, randomly selecting several pick-up and drop-off points in path 1, and the positions can be discontinuous. Then find the positions of the pick-up and drop-off points of path 1 selected in path 2, and then generate the offspring with the remaining pick-up and drop-off points in path 2. Finally, put the selected pick-up and drop-off points in path 1 into the remaining positions of the offspring in order. The Swap operator means randomly swapping two pick-up and drop-off points in a path sequence. The Reverse operator means randomly selecting a segment of pick-up and drop-off points in a path and reversing the pick-up and drop-off point segment.

[0140] As an alternative implementation, according to the offspring population C described in step S5 *, the first update of the POP is performed using the dominant archive update strategy, specifically including:

[0141] Determine whether the POP is empty to obtain a third judgment result;

[0142] If the third judgment result is yes, add the newly generated path planning scheme X new to the POP; the newly generated path planning scheme is an individual in the offspring population;

[0143] If the third judgment result is no, compare the newly generated path planning scheme X new with the existing path planning schemes in terms of dominance;

[0144] If there exists an existing path planning scheme Y that dominates X new or the existing path planning scheme Y is the same as X new , then discard X new ; if X new dominates the existing path planning scheme Y, then delete the dominated path planning scheme and add X new to the POP; if X new and the existing path planning scheme Y do not dominate each other, then add X new to the POP, and for the non-dominated path planning schemes in the POP, use the ε-dominance method for screening to obtain evenly distributed path planning schemes.

[0145] For the dominant archive update strategy mentioned in the above method, the following introduces this update strategy:

[0146] In the multi-objective intercity online car-hailing path planning problem, the comparison between path planning schemes is carried out through the multi-objective dominance relationship. The definition of the dominance relationship involved in this embodiment is as follows: For path planning schemes X and Y, if

[0147] 1) For all objective values, f j (X) ≤ f j (Y), j = 1, 2;

[0148] 2) There is at least one j such that f j (X) < f j (Y).

[0149] If both of the above two conditions are satisfied, then it is said that X dominates Y; otherwise, it is said that X and Y do not dominate each other, and X and Y are non-dominated solutions.

[0150] In addition, based on the multi-objective dominance relationship, this embodiment incorporates an ε-dominance strategy to limit the scale of the external archive during the search process. In the ε-dominance archive, each non-dominated solution in the archive corresponds to an associated vector B = {B1, B2, B3, B4, B5}, where B i = log(f i + 1) / log(1 + ε), and each non-dominated solution is stored in a hypercube. Therefore, the ε-dominance-based archive not only evenly distributes the non-dominated solutions but also limits the scale of the external archive during the search process.

[0151] S6. Perform a second update on all the planning schemes in the first POP through chain-based adaptive local search to obtain a second POP.

[0152] In this embodiment, chain-based adaptive local search is adopted to generate new path planning schemes. The adaptive operations involved in local search are defined by three operation operators, which define how to update and optimize the path planning scheme. At the same time, according to the roulette wheel strategy, the currently most potential task is selected. The roulette wheel strategy selects the most potential task in the current search process for local search to guide the optimal search and avoid resource waste. The mechanism of the adaptive local search chain selects potential solutions to achieve adaptive depth search. In the mechanism of the adaptive local search chain, the key lies in how to use multi-directional local search operations to construct an adaptive local search chain and how to select solutions during the local search chain process to further search the excellent candidate solution region obtained through knowledge transfer by local search operators and improve the quality of the solutions.

[0153] A multi-task optimization method for intercity online car-hailing path planning provided by the present invention defines the intercity online car-hailing path planning problem as a multi-objective problem with five objectives, which more comprehensively and realistically reflects the essence of the path planning problem; constructs an initial non-dominated planning scheme for the current path planning task through a series of path construction methods based on greed; then, based on the multi-task environment construction method of the problem decomposition strategy, decomposes a single path planning problem into multiple sub-problems representing different directions, and combines multiple sub-problems into multiple tasks to achieve multi-tasking and accelerate the process of generating path planning schemes; at the same time, in order to further utilize the synergistic effect existing between different solution regions, a knowledge transfer strategy is proposed to transfer the excellent path planning scheme solutions obtained from the solution regions represented by other different path planning tasks to the target task to optimize the diversity of the path planning scheme. Finally, chain-based adaptive local search is introduced to iteratively optimize the path planning scheme. The effective combination of these mechanisms can not only efficiently plan the travel path of intercity online car-hailing but also provide a set of high-quality planning schemes that meet different needs for intercity online car-hailing services using the multi-task multi-objective optimization method.

[0154] According to the actual situation of intercity online car-hailing travel services, the chain-based adaptive local search adopted is as follows:

[0155] The first local search: Randomly delete one order from each of the two paths selected from a path planning scheme, and then re-insert these two orders into the best positions of the path planning scheme;

[0156] The second local search: Delete all the orders of the two paths selected from a path planning scheme, and then re-insert them into the best positions of the path planning scheme;

[0157] The third local search: For the current path planning scheme, randomly select a path planning scheme different from itself from the external archive, exchange one path from each of these two schemes, then delete the orders that repeatedly appear on the exchanged paths in the current scheme, and re-insert the orders that do not appear in this scheme into the best positions;

[0158] According to the currently constructed multi-task environment, the roulette wheel strategy is defined as follows:

[0159] Initialize the selection probability prob of each task to 1, prob m = 1, m ∈ {1, …, K}. Select a task through the roulette wheel strategy, and reset the selection probability parameter of the currently selected task in real time. Where count m represents the number of times the solutions on all sub-problems of task T m enter the external archive in the local search chain, and invoke m represents the number of times the solutions of task T m perform local search on all sub-problems. Select the most promising task according to the current states of each task during the search process. This state refers to the ratio of the number of times the solutions on all sub-problems of the task enter the external archive in the local search chain to the number of local search times. The larger the ratio, the greater the probability that the task will be selected in the next round of roulette wheel.

[0160] Since the greedy sequence path construction method, the multi-task environment construction method based on the problem decomposition strategy, and the knowledge transfer method process first ensure the occupancy rate of the vehicle, the above local search will not change the number of paths in the original path planning scheme. The first local search and the second local search mainly generate new path planning schemes by mutating the original path planning scheme, while the third local search generates new path planning schemes by crossing different path planning schemes. Through the above definitions, the specific steps of the chain-based adaptive local search in this embodiment are as follows:

[0161] As an alternative implementation, in step S6, all path planning solutions in the first POP are updated for the second time through chain-based adaptive local search, which specifically includes:

[0162] S61. Set the number of times of adaptive local search Time to 0;

[0163] S62. Select the task T with the greatest current potential through the roulette wheel strategy m ; The task with the greatest potential is the task corresponding to the sub-problem with the greatest potential, and the sub-problem with the greatest potential is the sub-problem corresponding to the solution with the largest ratio of the number of times entering the POP in the local search chain to the number of local search times;

[0164] S63. Randomly adopt a local search strategy to perform local search operations on the solutions of each sub-problem of the task T m to obtain search solutions;

[0165] S64. Determine whether the search solution is better than the solution of the sub-problem. If so, use the search solution to replace the weight vector solution in the POP and continue to randomly adopt a local search strategy to perform local search operations on the solutions of the sub-problems; if not, return the solution before the local search of the current sub-problem;

[0166] Among them, determining whether the solution is better means determining whether the search solution x* obtained after the local search for the current sub-problem makes progress in the direction represented by the corresponding sub-problem (fob j (x*) is better than fob j (x), and the solution x will be replaced by x*).

[0167] S65. Increment the number of times of adaptive local search Time by 1, and determine whether Time is equal to 100. If it is equal, end; otherwise, return to the step "Select the task T with the greatest current potential through the roulette wheel strategy m ".

[0168] S7. According to the needs of the passengers in the order, select a path for the passengers from the path planning solutions in the second POP, allocate the corresponding vehicle and driver, and delete the allocated path from the second POP.

[0169] It should be noted that when allocating the path planning solution to the vehicle and driver in step S7, it needs to be allocated according to the preferences of frequent passengers. For example, when the passenger remarks in the order that they hope the path from the starting point to the terminal is the shortest, when allocating the vehicle and driver for the passengers of this order, the vehicle and driver corresponding to the path planning solution that minimizes f2 in the target optimization problem should be allocated.

[0170] S8. Set the POP set to an empty set and return to step S2.

[0171] The intercity online car-hailing service involved in this embodiment is mainly targeted at the road passenger transport market, combining the current Internet technology with the car rental industry, making up for the deficiencies of transitional public transportation in personalized travel, and playing an active role in optimizing the allocation of market resources and meeting the high-quality, personalized, and diversified travel needs of citizens. Therefore, efficiently and intelligently planning the travel routes of intercity online car-hailing can not only conduct vehicle scheduling in advance, bringing a comfortable travel experience to travel customers, but also saving transportation costs and bringing more benefits to the company and drivers.

[0172] To test the effectiveness of the proposed route planning method, the real order data of an intercity online car-hailing platform was used for testing.

[0173] The order situation in the test database is as follows: 1520 orders from City A to City B on a certain day, 4 different order passenger number requirements, and 2 types of time windows. Among them, the passenger number requirements are: 1, 2, 3, 4. The 2 types of time windows are: [6:30, 22:30] and [22:30, 6:30]. The first type of time window represents from 6:30 in the morning to 22:30 at night. Since most of the orders within this time window are during the day and are relatively safe, according to the preference, the weight coefficients are more preferred for the optimization of objective 2 and objective 3. The second type of time window represents from 22:30 at night to 6:30 in the morning. For the orders within this type of time window, more attention will be paid to safety, shortening the waiting time and delay time, and the number of on-call online car-hailing vehicles. Therefore, according to the preference, the weight coefficients are more preferred for the optimization of objective 1, objective 4, and objective 5.

[0174] By dynamically simulating 1520 orders and comparing them with the route planning scheme manually challenged by the company's professional dispatching customer service, the intercity online car-hailing travel route planning method based on the problem decomposition strategy multi-task multi-objective optimization algorithm proposed by the present invention has significantly decreased in all 5 objectives, and also has a significant improvement in the occupancy rate and passenger evaluation feedback.

[0175] In addition, to further verify the effectiveness of the proposed method, the comparative experimental results of the proposed method and the LSMOVRPTW algorithm were statistically analyzed. Two metrics widely used in multi-objective optimization algorithms, namely IGD and HV, were adopted. They were used to indicate the convergence and diversity of the algorithms from different perspectives. Among them, IGD and HV are unary metrics. The results show that the method of the present invention is significantly superior to the LSMOVRPTW algorithm in terms of both IGD and HV metrics. Specifically, in terms of the IGD metric, according to the single-problem analysis of the Wilcoxon test, the method of the present invention is significantly superior to the LSMOVRPTW algorithm in 40 test cases and inferior to the LSMOVRPTW algorithm in 1 test case. In terms of the HV metric, the method of the present invention is significantly superior to the LSMOVRPTW algorithm in 37 test cases and is slightly worse in 3 test cases. Based on the multi-problem analysis of the Wilcoxon test, the method of the present invention obtains R+ values higher than R- in terms of both IGD and HV metrics. It shows that in all rwMOVRPTW test cases, the method of the present invention is generally superior to the comparative algorithm. At the same time, in order to intuitively display the convergence and diversity of the method of the present invention and the LSMOVRPTW algorithm, the non-dominated solutions obtained in several representative test cases were mapped on the second objective and the third objective (f2 - f3), and the second objective and the fourth objective (f2 - f4). Compared with the comparative algorithm, the distribution of the non-dominated solutions of the method of the present invention is more extensive and closer to the Pareto front. In summary, the method proposed by the present invention can efficiently and intelligently handle the path planning problem of intercity online car-hailing trips.

[0176] Embodiment 2

[0177] This embodiment provides a multi-task optimization system for intercity online car-hailing path planning, including:

[0178] A model establishment module M1, configured to model the travel path planning problem according to the actual requirements of intercity online car-hailing travel path planning; the established model is a multi-objective optimization problem minf = {f1, f2, f3, f4, f5} including five objectives, where f1 is the number of online car-hailing vehicles, f2 is the total driving distance of online car-hailing vehicles to pick up and drop off passengers; f3 is the driving time of the longest driving path for online car-hailing vehicles to pick up and drop off passengers, f4 is the total waiting time of online car-hailing vehicles due to arriving at the pick-up and drop-off locations in advance, and f5 is the total delay time caused by online car-hailing vehicles arriving late.

[0179] An order judgment module M2, configured to judge whether there is an unprocessed online car-hailing order currently. If so, enter the step "generate an initial planning scheme for the order travel path through a greedy-based path construction method"; otherwise, enter the waiting state until a new order arrives and enter the step "generate an initial planning scheme for the order travel path through a greedy-based path construction method".

[0180] The initial planning module M3 is used to generate an initial planning scheme for the order travel path through a greedy-based path construction method, and save the initial planning scheme to the external archive solution set POP.

[0181] The problem decomposition module M4 is used to decompose the multi-objective optimization problem into N single-objective sub-problems according to the initial planning scheme by using a problem decomposition strategy, and combine the N sub-problems into K tasks.

[0182] The first update module M5 is used to set the k-th task among the K tasks as the target task T k , where k = 1, 2,..., K, and the remaining tasks are set as auxiliary tasks. The knowledge transfer method is used for optimization among the tasks to obtain a new path planning scheme, and the POP is first updated according to the new path planning scheme by using the ε-dominance archive update strategy to obtain the first POP.

[0183] The second update module M6 is used to perform a second update on all the planning schemes in the first POP through chain-based adaptive local search to obtain the second POP.

[0184] The scheme allocation module M7 is used to allocate each path planning scheme in the second POP to the vehicle and driver corresponding to the path, and delete the allocated path from the second POP.

[0185] The empty setting module M8 is used to set the POP set as an empty set and return to the step of "judging whether there is an unprocessed online car-hailing order currently".

[0186] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0187] Specific examples are applied in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A multi-task optimization method for intercity online car-hailing path planning, characterized in that, Including: S1. According to the actual requirements of the intercity online car-hailing travel path planning, model the travel path planning problem; the established model is a multi-objective optimization problem with five objectives minf = {f1, f2, f3, f4, f5}, where f1 is the number of online car-hailing vehicles, f2 is the total driving distance of the online car-hailing vehicles for picking up and dropping off passengers; f3 is the driving time of the longest driving path for the online car-hailing vehicles to pick up and drop off passengers, f4 is the total waiting time of the online car-hailing vehicles due to arriving at the pick-up and drop-off locations in advance, and f5 is the total delay time caused by the online car-hailing vehicles arriving late; S2. Judge whether there is an unprocessed online car-hailing order currently. If so, enter step S3; otherwise, enter the waiting state until a new order arrives and enter step S3; S3. Generate an initial planning scheme for the order travel path through a greedy-based path construction method, and save the initial planning scheme to the external archive solution set POP; S4. According to the initial planning scheme, decompose the multi-objective optimization problem into N single-objective sub-problems by using a problem decomposition strategy, and combine the N sub-problems into K tasks; S5. Designate the k-th task among the K tasks as the target task T k , where k = 1, 2,..., K, designate the remaining tasks as auxiliary tasks, optimize among the tasks using the knowledge transfer method to obtain a new path planning scheme, and perform the first update on the POP according to the new path planning scheme using the ε-dominance archive update strategy to obtain the first POP; Among them, the first update of the POP is performed using the ε-dominance archive update strategy according to the new path planning scheme, and the first POP is obtained, including: determining population C based on the target task and the auxiliary task, and randomly applying genetic operators to the population C to obtain the offspring population C * , and according to the offspring population C * , the first update of the POP is performed using the dominance archive update strategy, specifically including: Judge whether the POP is empty to obtain a third judgment result; If the result of the third judgment is yes, add the newly generated path planning scheme X new to the POP; the newly generated path planning scheme is an individual in the offspring population; If the result of the third determination is negative, the newly generated path planning scheme X new is compared with the existing path planning schemes for dominance; If there exists an existing path planning scheme Y that dominates X new or the existing path planning scheme Y is the same as X new then discard X new ; if X new dominates the existing path planning scheme Y, then delete the dominated path planning scheme and add X new to the POP; if X new and the existing path planning scheme Y do not dominate each other, then add X new to the POP, and for the non-dominating path planning schemes in the POP, use the ε-dominance method for screening to obtain uniformly distributed path planning schemes; S6. Perform a second update on all the planning schemes in the first POP through chain-based adaptive local search to obtain the second POP; S7. According to the needs of the passengers in the order, select a path from the path planning schemes of the second POP for the passengers and allocate the corresponding vehicles and drivers, and delete the allocated path from the second POP; S8. Set the POP set to an empty set and return to step S2.

2. The method according to claim 1, wherein The method of generating an initial planning scheme for the order travel path through a greedy-based path construction method and saving the initial planning scheme to the initial path planning solution set POP specifically includes: S31. Set the number of path planning schemes i' = 1; S32. Obtain all current orders, combine them into an order sequence, obtain the order attributes of all current said orders, and obtain a path planning scheme P i′ ; The order attributes include the number of people in the order, the starting point, the starting point service time, and the ending point; S33. Randomly select an order from the order sequence, denoted as the first order, create a path t according to the starting point and ending point of the first order, and use the starting point of the first order as the starting point of the online car-hailing vehicle p; the online car-hailing vehicle p is the one whose service time meets the starting point service time of the first order and is the closest to the starting point of the first order; S34. Judge whether the number of people in the first order is equal to the number of empty seats in the online car-hailing vehicle p; If so, enter step S310; if not, enter step S35; S35. Judge whether there are still unselected orders in the order sequence. If so, enter step S36, otherwise, enter step S310; S37. Randomly select an unselected order from the order sequence, denoted as the second order; S38. Determine whether the starting service time of the second order meets the service time of the online car-hailing p, obtaining a first judgment result, and determine whether adding the second order to the orders of the online car-hailing p meets five constraint conditions, obtaining a second judgment result; the five constraint conditions include: capacity constraint, time constraint, safety constraint, first service constraint, and second service constraint; the capacity constraint is that the number of passengers in each order cannot exceed the maximum passenger capacity of the online car-hailing, and the total number of passengers of all orders allocated on the same path cannot exceed the maximum passenger capacity of the online car-hailing; the safety constraint is that the online car-hailing driver needs to take a rest between two consecutive trips; the first service constraint is that the online car-hailing needs to let all passengers on the current path get off before picking up the next batch of passengers; the second service constraint is that in a path planning scheme, the pick-up and drop-off points of any order will only appear in one and the same path, and the pick-up point of each order must appear before the drop-off point of this path. If both the first judgment result and the second judgment result are yes, add the second order to the orders of the online car-hailing p; otherwise, return to step S35. S39. Determine whether the number of passengers in the orders currently added to the online car-hailing p is equal to the number of empty seats in the online car-hailing p. If yes, enter step S310; if not, return to step S35. S310. Use the greedy strategy to sort the destinations of all orders of the online car-hailing p and the origins of the remaining orders except the first order. According to the sorting result, insert the origins and destinations of the remaining orders of the online car-hailing p into the path t to obtain the path planning route of the online car-hailing p, and add the path planning route to the path planning scheme P i′ In this case, set the status of all orders of the online car-hailing p to "processed", and remove them from the order sequence; S311. Add the path planning scheme P i′ to the initial path planning solution set POP; S312. Determine whether there is an unprocessed order in the order sequence. If yes, return to step S33; otherwise, enter step S313. S313. Determine whether the number of path planning schemes in the POP reaches N. If yes, end; if not, execute i′ = i′ + 1 for i′ and return to step S32.

3. The method according to claim 1, wherein The method of decomposing the multi-objective optimization problem into N single-objective sub-problems according to the initial planning scheme and combining the N sub-problems into K tasks specifically includes: Decompose the multi-objective optimization problem into N sub-problems by using the boundary crossover method, form a sub-problem set with the N sub-problems, and initialize N solutions according to the initial path planning scheme to obtain an initial planning solution set; the N solutions correspond to the N sub-problems one by one. Randomly select an unselected sub-problem from the N sub-problems, denoted as the first sub-problem, and select the solution with the smallest fitness value in the initial planning solution set corresponding to the first sub-problem as the weight vector solution of the first sub-problem, remove the weight vector solution from the initial planning solution set, and return to the step "Randomly select an unselected sub-problem from the N sub-problems" until the initial planning solution set is empty. Initialize the number of multi-tasks K, and evenly divide the N sub-problems into K tasks. Calculate the center point x of each task, determine the central sub-problem according to the center point, add the central sub-problem to the corresponding task, and remove the central sub-problem from the sub-problem set. Calculate the n-1 points closest to each said center point x using the Euclidean distance, and determine the sub-problems corresponding to the n-1 points, denoted as the nearest sub-problems. Add the nearest sub-problems to the task corresponding to the center point x, and remove the nearest sub-problems from the set of sub-problems.

4. The method according to claim 3, characterized in that, Optimize among the tasks using a knowledge transfer method to obtain a new path planning scheme. Perform the first update on the POP using the ε-dominance archive update strategy according to the new path planning scheme, specifically including: S51. Encoding and population initialization; S52. Set the population C to be empty, and add the weight vector solution of the target task T k to the population C, where k = 1; S53. Use the NSGA-II algorithm to perform fast non-dominated sorting on the weight vector solutions of each task in the auxiliary tasks, and obtain the first front solutions according to the sorting results; S54. Combine the weight vector solutions of the target task and the first front solutions to obtain a combined path planning scheme, and use the combined path planning scheme as the parent population; S55. Determine whether the number of the first front solutions is greater than N / K; if so, go to step S56; if not, add the first front solutions to the population C; S56. Randomly select N / K solutions from the first front solutions and add them to the population C; Randomly apply genetic operators to the population C to obtain the offspring population C * , and based on the offspring population C * , perform the first update on the POP using the dominance archive update strategy; Combine the parental population C and the offspring population C * to obtain the combined population C 联 , and delete the duplicate path planning schemes in the combined population C 联 ; S59. Use an environmental selection operation for combined population C 联 Re-select the solution with the minimum fitness value as the weight vector solution for the sub-problem corresponding to the target task described in S510. Determine whether k is equal to K; if not, perform the operation k = k + 1 on k and return to step S53; if so, end.

5. The method according to claim 1, wherein The dominance comparison is as follows: Determine the dominance relationship between path planning scheme X new and Y: Determine path planning scheme X new and Y simultaneously satisfy condition 1 and condition 2; condition 1 is that for all target values f j (X new ) ≤ f j (Y), j = 1, 2; condition 2 is that there exists at least one j such that f j (X new ) < f j (Y); If so, then X is said to new dominate Y; Otherwise, it is said that X new and Y do not dominate each other.

6. The method according to claim 4, wherein Perform a second update on all path planning schemes in the first POP through chain-based adaptive local search, specifically including: Set the number of times of adaptive local search Time to 0; Select the current task T with the greatest potential through the roulette wheel strategy m ; The task with the greatest potential is the task corresponding to the sub-problem with the greatest potential, and the sub-problem with the greatest potential is the sub-problem corresponding to the solution with the largest ratio of the number of times entering the POP to the number of local search times in the local search chain; Randomly adopt a local search strategy to perform local search operations on the solutions of each sub-problem of the task T m to obtain search solutions; the local search operations include the first local search, the second local search, and the third local search; the first local search is to randomly delete one order from each of the two paths selected from a path planning scheme, and then insert the two deleted orders into the best position of the path planning scheme; the best position is defined by the genetic operator; the second local search is to delete all the orders of the two paths selected from a path planning scheme, and then re-insert all the orders into the best position of the path planning scheme; the third local search is for the current path planning scheme, randomly select another path planning scheme from the first POP, exchange one path selected from the current path planning scheme with one path selected from the other path planning scheme to obtain a cross path planning scheme, delete the orders that appear repeatedly in the cross path planning scheme from the current path planning scheme, and re-insert the orders that do not appear in the cross path planning scheme into the best position; Determine whether the search solution is better than the solution of the sub-problem; if so, replace the weight vector solution with the search solution in the POP, and continue to randomly use a local search strategy to perform local search operations on the solution of the sub-problem; if not, return to the solution before the local search of the current sub-problem; Increment the number of times Time of the adaptive local search by 1, and determine whether Time is equal to 100. If it is equal, end; otherwise, return to the step "Select the task T with the greatest current potential through the roulette wheel strategy". m ".

7. A multi-task optimization system for intercity online car-hailing route planning, which is used to execute a multi-task optimization method for intercity online car-hailing route planning as described in claim 1, characterized in that, Including: A model establishment module, which is used to model the travel path planning problem according to the actual requirements of intercity online car-hailing travel path planning; the established model is a multi-objective optimization problem with five objectives min f = {f1, f2, f3, f4, f5}, where f1 is the number of online car-hailing vehicles, f2 is the total driving distance of online car-hailing vehicles to pick up and drop off passengers; f3 is the driving time of the longest driving path for online car-hailing vehicles to pick up and drop off passengers, f4 is the total waiting time of online car-hailing vehicles due to arriving at the pick-up and drop-off locations in advance, and f5 is the total delay time caused by online car-hailing vehicles arriving late; An order judgment module, which is used to judge whether there is an unprocessed online car-hailing order currently. If there is, enter the step "Generate an initial planning scheme for the order travel path through a greedy path construction method"; otherwise, enter the waiting state until a new order arrives, and enter the step "Generate an initial planning scheme for the order travel path through a greedy path construction method"; An initial planning module, which is used to generate an initial planning scheme for the order travel path through a greedy path construction method, and save the initial planning scheme to the external archive solution set POP; A problem decomposition module, configured to decompose the multi-objective optimization problem into N single-objective sub-problems by adopting a problem decomposition strategy according to the initial planning scheme, and combine the N sub-problems into K tasks; The first update module is used to set the k-th task among the K tasks as the target task T k , where k = 1, 2,..., K, and the remaining tasks are set as auxiliary tasks. The knowledge transfer method is used to optimize among the tasks to obtain a new path planning scheme. According to the new path planning scheme, the ε-dominance archive update strategy is used to perform the first update on the POP to obtain the first POP; A second update module, configured to perform a second update on all the planning schemes in the first POP through chain-based adaptive local search to obtain a second POP; A scheme allocation module, configured to allocate each path planning scheme in the second POP to the vehicle and driver corresponding to the path, and delete the allocated path from the second POP; An empty setting module, configured to set the POP set as an empty set, and return to the step of "judging whether there is an unprocessed online car-hailing order currently".

Images