Electric network car-hailing online car-sharing scheduling method and system under dual-network integration

Abstract

The invention discloses an electric network car-hailing online car-sharing scheduling method and system under dual-network integration, and the method comprises the following steps: obtaining region information at different times in a preset region according to historical data, and carrying out the sampling of the region information to generate training data; according to the regional information, carrying out modeling on a dual-network integration electric network car-hailing online car-sharing scheduling problem, determining an optimization target of a model, and constructing a target function and a constraint condition; according to the optimization target and the training data, programming a co-evolution charging rule pool and an order receiving rule pool by adopting a gene expression; and taking the charging rule pool and the order receiving rule pool as action spaces of a deep reinforcement learning hyper-heuristic algorithm, solving the model based on the deep reinforcement learning hyper-heuristic algorithm, selecting actions from the action spaces for idle vehicles, and updating states of the vehicles according to the selected actions. According to the method, the convergence speed is improved through co-evolution of the charging rule and the order receiving rule population, and meanwhile, the robustness is improved by taking the rule pool as an action space of deep reinforcement learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle scheduling, in particular to an online carpooling scheduling method and system for electric online car-hailing under dual-network fusion. BACKGROUND

[0002] At present, most online car-hailing scheduling systems mainly consider the operation characteristics of traditional fuel vehicles when designing and optimizing, and rarely optimize for the special needs of electric vehicles. As online car-hailing, electric vehicles have limited endurance, long charging time, uneven distribution of charging facilities, and fluctuating electricity prices with charging stations and time.

[0003] Carpooling integrates the travel needs of multiple passengers, reducing the empty running rate of vehicles, which not only helps to alleviate traffic congestion, but also reduces energy consumption and environmental pollution. It is of great significance to research the integration of road network and power grid, include electric vehicles in online car-hailing fleets, optimize order execution sequence, and develop a detailed vehicle charging strategy. SUMMARY

[0004] In order to overcome the defects and deficiencies of the prior art, the present application provides an online carpooling scheduling method and system for electric online car-hailing under dual-network fusion, which establishes an online carpooling scheduling model for electric online car-hailing from the perspective of the overall operation of the operator, improves the convergence speed by co-evolving the charging rule and order-accepting rule populations, and improves the robustness by using the rule pool as the action space of deep reinforcement learning.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] The present application provides an online carpooling scheduling method for electric online car-hailing under dual-network fusion, comprising the following steps:

[0007] According to historical data, obtain regional information at different times in a predetermined area, and sample to generate training data;

[0008] Model the online carpooling scheduling problem of dual-network fusion electric online car-hailing according to the regional information, determine the optimization objective of the model, and construct the objective function and constraint conditions;

[0009] According to the optimization objective and training data, use gene expression programming to co-evolve the charging rule pool and order-accepting rule pool;

[0010] Use the charging rule pool and order-accepting rule pool as the action space of the deep reinforcement learning super-heuristic algorithm, solve the model based on the deep reinforcement learning super-heuristic algorithm, select actions from the action space for idle vehicles, and update the state of the vehicle according to the selected actions.

[0011] As a preferred technical solution, the area information at different times includes road traffic costs at different times, charging station charging prices at different times, and order occurrence regularities at different times.

[0012] As a preferred technical solution, the objective function is expressed as:

[0013]

[0014] Wherein, TD represents the total distance of vehicle travel, CC represents the total charging cost, OT represents the overtime, PT represents the penalty time, RA represents the number of abandoned orders, MD is the maximum travel distance, MC is the maximum charging cost, RN is the total number of orders, MT is the maximum overtime allowed for the order, and p1, p2, and p3 are preset weight parameters.

[0015] As a preferred technical solution, the constraint conditions include: passenger number constraint, service time constraint, power constraint, and order service constraint.

[0016] As a preferred technical solution, the gene expression programming is used to cooperate evolution of the charging rule pool and the order accepting rule pool, which specifically includes:

[0017] The population and pool of charging rules and order accepting rules are initialized respectively;

[0018] The order accepting rule population is evaluated, and individuals in the charging rule pool are randomly selected for cooperative evaluation;

[0019] The phenotype repeatability detection is used to select individuals from the order accepting rule population to update the order accepting rule pool;

[0020] The replicas of individuals in the charging rule pool are migrated to the order accepting rule population, and the order accepting rule offspring population is optimized through crossover and mutation operations;

[0021] The charging rule population is evaluated, and individuals in the order accepting rule pool are randomly selected for cooperative evaluation;

[0022] The phenotype repeatability detection is used to select individuals from the charging rule population to update the charging rule pool;

[0023] The replicas of individuals in the order accepting rule pool are migrated to the charging rule population, and the charging rule offspring population is optimized through crossover and mutation operations;

[0024] If the termination condition is reached, the updated order accepting rule pool and charging rule pool are output, otherwise the evaluation of the order accepting rule population is returned.

[0025] As a preferred technical solution, the phenotype repeatability detection specifically includes:

[0026] Decision points are randomly selected from the training data as test decision points;

[0027] According to the fitness value of the individual in the population, the weight of the excellent individual is allocated, and the weight calculation formula is represented as:

[0028]

[0029] Wherein, w i represents the weight of the i th individual, x i represents the fitness value of the i th individual, and N represents the number of excellent individuals.

[0030] According to the excellent individual, the candidate decision of the test decision point is weighted and sorted, and the calculation formula is:

[0031]

[0032] Wherein, r m，n is the weighted sorting of the n th candidate decision in the m th test decision point, is the sorting number of the n th candidate decision in the m th test decision point according to the j th individual.

[0033] According to the ascending order of fitness, if the decision of the individual at any test decision point is located in the weighted sorting ranking of the set proportion, and the decision of the individual at the test decision point is not consistent with the decision of the individual in the pool, then the individual is added to the pool.

[0034] If there are individuals in the pool in the set proportion of the weighted sorting ranking in all test decision points, or the population traversal is completed and the construction of the pool is not completed, then the tournament selection is carried out in the population, and the construction of the pool is continued.

[0035] As a preferred technical scheme, the model is solved based on the deep reinforcement learning super heuristic algorithm, the action is selected from the action space for the idle vehicle, and the state of the vehicle is updated according to the selected action, which specifically includes:

[0036] Obtain the order quantity and working state of the current time;

[0037] Traverse all the current idle vehicles, select the next action from the action space for the idle vehicle, and update the state of the vehicle according to the selected action;

[0038] If the idle is selected, the next time scheduling is waited;

[0039] If the charging rule is selected, the charging station is selected for charging;

[0040] If the order receiving rule is selected, the next order node to be visited is selected, and the order node includes a new passenger pickup point meeting the carpooling demand, or a corresponding passenger drop-off point of the visited passenger pickup point.

[0041] The application also provides a double-network fusion electric network car-hailing online carpooling dispatching system for realizing the double-network fusion electric network car-hailing online carpooling dispatching method.

[0042] The training data generation module is configured to acquire regional information at different times in a preset region according to historical data, and sample and generate training data.

[0043] The problem modeling module is configured to model the double-network fusion electric network car-hailing online carpooling dispatching problem according to the regional information, determine an optimization target of the model, and construct a target function and constraint conditions.

[0044] The cooperative evolution module is configured to use gene expression programming to cooperatively evolve a charging rule pool and an order-accepting rule pool according to the optimization target and the training data.

[0045] The model solving module is configured to take the charging rule pool and the order-accepting rule pool as an action space of a deep reinforcement learning hyper-heuristic algorithm, solve the model based on the deep reinforcement learning hyper-heuristic algorithm, select an action for an idle vehicle from the action space, and update a state of the vehicle according to the selected action.

[0046] The application also provides a computer-readable storage medium storing a program, which, when executed by a processor, realizes the double-network fusion electric network car-hailing online carpooling dispatching method.

[0047] The application also provides a computer device including a processor and a memory for storing a program executable by the processor, and the processor realizes the double-network fusion electric network car-hailing online carpooling dispatching method when executing the program stored in the memory.

[0048] Compared with the prior art, the application has the following advantages and beneficial effects:

[0049] (1) The application establishes an electric network car-hailing online carpooling dispatching model from the perspective of an operator as a whole, fully considers problems such as limited endurance, long charging time, non-widespread distribution of charging facilities, and fluctuation of electricity prices with charging stations and time when an electric vehicle is used as a network car service, and considers carpooling travel demand, thereby reducing the empty running rate of vehicles by integrating travel demands of multiple passengers.

[0050] (2) The application proposes a cooperative evolution charging rule and order-accepting rule population, improves the convergence speed, avoids the shortcoming of excessive coupling of charging rules and order-accepting rules by cooperatively evolving the charging rule and order-accepting rule population, and makes the system more flexible to cope with different operation scenarios, thereby improving the flexibility and adaptability of dispatching.

[0051] (3) The present invention adopts a deep reinforcement learning hyper-heuristic algorithm based on genetic expression programming, and uses the rule pool as the action space of deep reinforcement learning. It can select different rules in the rule pool according to different environments to achieve complementarity. This not only improves the robustness of the method, but also enables the system to adaptively adjust strategies when facing complex and changeable traffic and order demands, and optimize the overall operational efficiency online. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the process of online carpooling scheduling method for electric online car-hailing under dual-network integration of the present invention;

[0053] Figure 2 Schematic diagram of the implementation process of the collaborative evolution based on gene expression programming of the present invention;

[0054] Figure 3 This is a schematic diagram of the process of optimizing the deep reinforcement learning hyper-heuristic algorithm of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] Example 1

[0057] like Figure 1 As shown, this embodiment provides an online carpooling scheduling method for electric online ride-hailing vehicles under dual-network integration. In the process of order dynamics, the driving routes of a group of vehicles are efficiently planned online to meet power constraints and take into account charging conditions, while meeting a series of online ride-hailing order requirements and minimizing total costs. In this process, factors such as the total driving distance of the fleet, total charging costs, order timeouts, and order completion rates are comprehensively considered. This embodiment adopts a deep reinforcement learning hyper-heuristic algorithm based on gene expression programming. A set of low-level heuristic rules generated by gene expression programming is used as the action space of the deep reinforcement learning agent. The deep reinforcement learning agent acts as a selector and can make intelligent choices from a large number of low-level heuristic rules according to specific situations in different scenarios, thereby achieving efficient optimization. Specifically, the following steps are included:

[0058] S1: Based on historical data, obtain regional information at different times within the preset area and sample to generate training data;

[0059] In this embodiment, regional information at different times includes road travel costs at different times, charging station prices at different times, and the occurrence patterns of orders at different times. The collected regional information is randomly sampled to generate training data.

[0060] S2: According to the area information, the double-network fusion electric network car online carpooling scheduling problem is modeled, and the optimization objective of the model is determined;

[0061] In this embodiment, a target function considering the overall operation is constructed, which comprehensively considers the total driving distance of the vehicle fleet, the total charging cost, the order overtime condition and the order completion rate, so as to realize the optimization of the overall operation efficiency, and the target function is expressed as:

[0062]

[0063] The optimization objectives include: total distance TD of vehicle driving, total charging cost CC, overtime OT, penalty time PT and abandoned order number RA, wherein MD is the maximum driving distance, MC is the maximum charging cost, RN is the total number of orders, MT is the maximum overtime allowed for the order, which is different according to different training cases, and p1, p2 and p3 are weight parameters defined in advance, which are used to balance the relative importance between the optimization objectives, so as to ensure that the target function can comprehensively reflect the key indicators in the operation process;

[0064] In this embodiment, the constraint conditions of the model are determined to ensure the feasibility of the electric network car online carpooling scheduling, and the specific constraint conditions are as follows:

[0065] (1) Passenger number constraint: the number of passengers carried by each vehicle does not exceed its maximum passenger capacity, so as to ensure the comfort of passengers and the safety of vehicles;

[0066] (2) Service time constraint: the maximum service time of each vehicle cannot exceed the maximum service time set in advance, so as to avoid excessive fatigue operation of the vehicle and ensure the operation efficiency;

[0067] (3) Power constraint: the power of each vehicle can meet the driving demand, so as to ensure that the vehicle will not interrupt the service due to insufficient power during the task execution;

[0068] (4) Order service constraint: each order has a clear pickup point and drop-off point, the vehicle must first arrive at the pickup point and then arrive at the drop-off point, and each order must be served by the same vehicle, and the pickup point and drop-off point of each order can only be visited once in the whole scheduling process, so as to ensure the integrity and continuity of the order service;

[0069] S3: According to the optimization objective and the training data, the gene expression programming is used to cooperate with the charging rule pool and the order receiving rule pool, the gene expression programming is a symbolic regression method based on evolutionary algorithm, which combines the advantages of genetic algorithm and genetic programming, uses linear chromosome coding with fixed length to encode genetic information, and generates expression trees with different sizes and shapes through gene expression system.

[0070] The core features of GP include:

[0071] Genes and chromosomes: Each chromosome consists of multiple genes, and each gene consists of a head and a tail. The head contains functions and terminal symbols, while the tail only contains terminal symbols.

[0072] Gene expression: Chromosomes are converted into expression trees by the gene expression system, which can represent complex mathematical relationships or program logic.

[0073] Genetic operations: Optimize chromosomes through genetic operations such as selection, crossover, mutation, etc. These operations include gene recombination, single-point recombination, two-point recombination, shift operation, etc.

[0074] In this embodiment, the process of co-evolution specifically includes:

[0075] (1) Initialize the population and pool of charging rules and order-accepting rules, respectively;

[0076] (2) Evaluate the order-accepting rule population, which requires randomly selecting individuals from the charging rule pool for co-evaluation;

[0077] (3) Update the order-accepting rule pool by selecting individuals from the order-accepting rule population using phenotype repeatability detection;

[0078] (4) Migrate the replicas of individuals in the charging rule pool to the order-accepting rule population, and optimize the order-accepting rule offspring population through crossover and mutation operations;

[0079] (5) Evaluate the charging rule population, which requires randomly selecting individuals from the order-accepting rule pool for co-evaluation;

[0080] (6) Update the charging rule pool by selecting individuals from the charging rule population using phenotype repeatability detection;

[0081] (7) Migrate the replicas of individuals in the order-accepting rule pool to the charging rule population, and optimize the charging rule offspring population through crossover and mutation operations;

[0082] (8) If the termination condition is met, output the updated order-accepting rule pool and charging rule pool, otherwise, go to step (2).

[0083] For example, Figure 2As shown, in the co-evolution process based on gene expression programming, firstly, two populations and corresponding pools are initialized respectively, wherein the pool stores the replicas of the corresponding excellent population individuals. Then, the order receiving rules population is evaluated. Since the order receiving rules and the charging rules need to be used simultaneously during the evaluation, multiple individuals are randomly sampled from the charging rules pool for evaluation, and the final fitness is the average value. The specific number of samples can be determined according to the complexity of the case. The purpose of this is to reduce the coupling degree between the order receiving rules and the charging rules, thereby enhancing the robustness of the system. Then, the repeatability detection operation is used to update the order receiving rules pool. After the order receiving rules pool is updated, the replicas of the individuals in the charging rules pool are migrated to the order receiving rules population, and the genetic operation is used for co-evolution to optimize the order receiving rules offspring population. Then, the order receiving rules are evaluated similarly, and the update and evolution operations are performed to complete one round of iteration. If the termination condition is reached, the updated order receiving rules pool and charging rules pool are output; otherwise, the next round of iteration is entered.

[0084] In this embodiment, the phenotype repeatability detection operation specifically includes:

[0085] Randomly extracting decision points from the training data as test decision points;

[0086] According to the fitness values of the individuals in the population, the top one-third excellent individuals are assigned weights, and the weight calculation formula is:

[0087]

[0088] In the formula, w i represents the weight of the i-th individual, x i represents the fitness value of the i-th individual, and N represents the number of the top one-third excellent individuals.

[0089] According to the top one-third excellent individuals, the candidate decisions of the test decision points are weighted and sorted, and the calculation formula is:

[0090]

[0091] In the formula, r m,n is the weighted sorting of the n-th candidate decision in the m-th test decision point, is the sorting number of the n-th candidate decision in the m-th test decision point according to the j-th individual;

[0092] The population is traversed in ascending order of fitness, and if the decision made by the individual at any test decision point is located in the top one-third weighted sorting ranking and is not consistent with the decision made by the individual in the pool at the test decision point, the individual is added to the pool.

[0093] If the top third of the weighted ranking pool of all test decision points have individuals selected, or the population is traversed, one of the above two cases occurs and the pool construction is not completed, then the tournament selection is performed in the population, and the pool construction is continued. By using the phenotype-based repeatability detection, the complementarity of each individual in the pool can be enhanced by increasing the diversity of individual performance in the pool.

[0094] S4: According to the charging rule pool and the order receiving rule pool, the deep reinforcement learning super heuristic algorithm is used for optimization.

[0095] In the embodiment, the model is solved according to the charging rule pool and the order receiving rule pool as the action space of the deep reinforcement learning super heuristic algorithm, the rules are flexibly selected according to different environments by the deep reinforcement learning super heuristic algorithm, and the rules have good adaptability and can fully play the complementarity of different rules, specifically including:

[0096] Obtain the order quantity and working state of the current time;

[0097] Traverse all current idle vehicles, and the idle vehicle refers to a vehicle that has arrived at a specified location, and the specified location includes a pickup point, a drop-off point and a charging station after charging is completed;

[0098] Select the next action for the idle vehicle from the action space of the deep reinforcement learning, and the action space includes the charging rule pool, the order receiving rule pool and the idle state;

[0099] Update the state of the vehicle according to the selected action: if the idle state is selected, wait for the next time scheduling; if the charging rule is selected, select a charging station for charging; if the order receiving rule is selected, select the next order node to be visited, and the order node can be a new pickup point meeting the carpooling demand or a corresponding drop-off point of the visited pickup point;

[0100] Scheduling is performed once every fixed time period, and the time period is determined according to the system busy degree.

[0101] As shown in Figure 3 According to the state of the environment, a suitable virtual action is selected by the deep reinforcement learning super heuristic algorithm based on the deep neural network, and the virtual action subsequently selects a low-level heuristic rule. After the execution of the heuristic rule, the actual action changes the environment, thereby causing the environment to change to a new state. Through step-by-step iteration, the algorithm ultimately completes the optimization.

[0102] The present application establishes an online carpooling dispatching model from the global perspective of the operator, improves the convergence speed by means of the cooperative evolution of the charging rule and the order receiving rule population, and improves the robustness by taking the rule pool as the action space of the deep reinforcement learning.

[0103] Embodiment 2

[0104] The embodiment provides an online carpooling dispatching system for electric network ride-hailing under dual-network fusion, which is used for implementing the online carpooling dispatching method for electric network ride-hailing under dual-network fusion in the embodiment 1, and comprises a training data generation module, a problem modeling module, a co-evolution module and a model solving module.

[0105] In the embodiment, the training data generation module is used for obtaining regional information at different times in a preset area according to historical data, and sampling and generating training data.

[0106] In the embodiment, the problem modeling module is used for modeling the online carpooling dispatching problem for dual-network fusion electric network ride-hailing according to the regional information, determining an optimization target of the model, and constructing a target function and a constraint condition.

[0107] In the embodiment, the co-evolution module is used for adopting gene expression programming to co-evolve a charging rule pool and an order accepting rule pool according to the optimization target and the training data.

[0108] In the embodiment, the model solving module is used for taking the charging rule pool and the order accepting rule pool as an action space of a deep reinforcement learning super-heuristic algorithm, solving the model based on the deep reinforcement learning super-heuristic algorithm, selecting an action for an idle vehicle from the action space, and updating a state of the vehicle according to the selected action.

[0109] Embodiment 3

[0110] The embodiment provides a storage medium, which can be a ROM, a RAM, a magnetic disk, an optical disk or the like storage medium. The storage medium stores one or more programs. When the programs are executed by a processor, the online carpooling dispatching method for electric network ride-hailing under dual-network fusion in the embodiment 1 is implemented.

[0111] Embodiment 4

[0112] The embodiment provides a computing device, which can be a desktop computer, a notebook computer, a smart phone, a PDA handheld terminal, a tablet computer or other terminal device with a display function. The computing device comprises a processor and a memory. The memory stores one or more programs. When the processor executes the programs stored in the memory, the online carpooling dispatching method for electric network ride-hailing under dual-network fusion in the embodiment 1 is implemented.

[0113] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principle of the present application shall be equivalent replacement modes and shall be included in the protection scope of the present application.

Claims

1. A method for online carpooling scheduling of electric online taxis under dual-network integration, characterized in that: The steps include: Based on historical data, obtain regional information at different times within the preset area and sample to generate training data; Model the online carpooling scheduling problem of electric ride-hailing vehicles with dual-network integration based on regional information, determine the optimization goal of the model, and construct the objective function and constraints; Based on the optimization objectives and training data, gene expression programming is used to co-evolve the charging rule pool and the order-taking rule pool; The charging rule pool and the order acceptance rule pool are used as the action space of the deep reinforcement learning hyper-heuristic algorithm. The model is solved based on the deep reinforcement learning hyper-heuristic algorithm, and actions are selected from the action space for idle vehicles. The vehicle status is updated according to the selected actions.

2. The online carpooling scheduling method for electric online taxis under dual-network integration according to claim 1 is characterized in that: Regional information at different times includes road travel costs at different times, charging prices at charging stations at different times, and the occurrence patterns of orders at different times.

3. The online carpooling scheduling method for electric online taxis under dual-network integration according to claim 1 is characterized in that: The objective function is expressed as: Among them, TD represents the total distance traveled by the vehicle, CC represents the total charging cost, OT represents the timeout period, PT represents the penalty period, RA represents the number of abandoned orders, MD represents the maximum driving distance, MC represents the maximum charging fee, RN represents the total number of orders, MT represents the maximum timeout period allowed for the order, and p1, p2, and p3 represent the preset weight parameters.

4. The online carpooling scheduling method for electric online taxis under dual-network integration according to claim 1 is characterized in that: Constraints include: passenger number constraints, service time constraints, power constraints and order service constraints.

5. The online carpooling scheduling method for electric online taxis under dual-network integration according to claim 1 is characterized in that: Gene expression programming is used to co-evolve the charging rule pool and the order-taking rule pool, specifically including: Initialize the population and pool of charging rules and order-taking rules respectively; Evaluate the order-taking rule population and randomly select individuals from the charging rule pool for collaborative evaluation; Individuals are selected from the order-taking rule population using phenotypic repeatability testing to update the order-taking rule pool; Migrate the replicas of individuals in the charging rule pool to the order-taking rule population, and co-evolve the optimized order-taking rule offspring population through crossover and mutation operations; Evaluate the charging rule population and randomly select individuals from the order-taking rule pool for collaborative evaluation; Phenotypic repeatability detection is used to select individuals from the charging rule population to update the charging rule pool; Migrate the replicas of individuals in the order-taking rule pool to the charging rule population, and co-evolve and optimize the charging rule offspring population through crossover and mutation operations; If the termination condition is met, the updated order-taking rule pool and charging rule pool are output; otherwise, the order-taking rule population is returned for evaluation.

6. The method for online carpooling scheduling of electric online taxis under dual-network integration according to claim 5 is characterized in that: Phenotypic reproducibility testing specifically includes: Randomly select decision points from the training data as test decision points; According to the fitness value of individuals in the population, weights are assigned to excellent individuals. The weight calculation formula is expressed as: Among them, w i represents the weight of the i-th individual, x i represents the fitness value of the i-th individual, and N represents the number of excellent individuals; According to the excellent individuals, the candidate decisions for the test decision point are weighted and ranked. The calculation formula is: Among them, r m，n is the weighted ranking of the nth candidate decision in the mth test decision point, The ranking number of the nth candidate decision according to the jth individual in the mth test decision point; Traverse the population in ascending order of fitness. If the decision made by an individual at any test decision point is in the weighted ranking of the set proportion and is not consistent with the decision made by the individual in the pool at the test decision point, then add it to the pool; If all test decision points have individuals selected in the set proportion weighted ranking pool, or the population traversal is completed and the pool construction is not complete, a tournament selection is performed in the population to continue to complete the pool construction.

7. The method for online carpooling scheduling of electric online taxis under dual-network integration according to claim 1 is characterized in that: The model is solved based on a deep reinforcement learning hyper-heuristic algorithm, and an action is selected from the action space for the idle vehicle. The vehicle state is updated based on the selected action, including: Get the current order quantity and work status; Traverse all currently idle vehicles, select the next action for the idle vehicle from the action space, and update the vehicle state according to the selected action; If you choose Idle, wait for the next time to be scheduled; If you select the charging rule, you will select a charging station for charging; If you select an order acceptance rule, select the order node to be visited next. The order node includes a new pick-up point that meets the carpooling needs, or a corresponding drop-off point for a previously visited pick-up point.

8. An online carpooling dispatching system for electric online taxis under dual-network integration, characterized in that: The method for online carpooling scheduling of electric online taxis under dual-network integration according to any one of claims 1 to 7 comprises: a training data generation module, a problem modeling module, a co-evolution module, and a model solving module; The training data generation module is used to obtain regional information at different times within a preset area based on historical data, and to generate training data by sampling; The problem modeling module is used to model the online carpooling scheduling problem of the dual-network integrated electric network-hailing vehicle based on regional information, determine the optimization goal of the model, and construct the objective function and constraint conditions; The co-evolution module is used to co-evolve the charging rule pool and the order-taking rule pool using gene expression programming according to the optimization target and training data; The model solving module is used to use the charging rule pool and the order acceptance rule pool as the action space of the deep reinforcement learning hyper-heuristic algorithm, solve the model based on the deep reinforcement learning hyper-heuristic algorithm, select actions from the action space for idle vehicles, and update the vehicle status according to the selected actions.

9. A computer-readable storage medium storing a program, characterized in that: When the program is executed by the processor, the online carpooling scheduling method for electric online taxis under dual-network integration is implemented as described in any one of claims 1-7.

10. A computer device comprising a processor and a memory for storing a program executable by the processor, characterized in that: When the processor executes the program stored in the memory, it implements the online carpooling scheduling method for electric online taxis under dual-network integration as described in any one of claims 1-7.

Images