Logistics goods taking and delivering path optimization method and system

By modeling the vehicle path problem as a multi-objective optimization problem and optimizing it with evolutionary algorithms and deep reinforcement learning models, the problem that VRPSPDT is prone to fall into local optimality in the existing technology is solved, and more efficient and diverse logistics pick-and-delivery path optimization is achieved.

CN120106325APending Publication Date: 2025-06-06SHENZHEN UNIV
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202510048037.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-01-13
Publication Date
2025-06-06

AI Technical Summary

Technical Problem

The prior art is prone to falling into local optimum when solving the problem of simultaneous pick-up and delivery of goods and vehicle path problems with time windows (VRPSPDT), which are difficult to maintain the diversity of solutions and ignore long-term returns.

Method used

By constructing a complete graph, the vehicle path problem is modeled as a multi-objective optimization problem and decomposed into multiple sub-problems. The sequential crossover and random variation in the evolutionary algorithm are used to solve multiple iteratively in combination with the deep reinforcement learning (DRL) model to optimize the logistics pick-up and delivery path.

Benefits of technology

It improves the diversity and quality of logistics pick-up and delivery paths, enhances the exploration ability and overall performance of understanding, and effectively avoids local optimal traps.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120106325A_ABST
    Figure CN120106325A_ABST
Patent Text Reader

Abstract

The invention provides a logistics pickup and delivery path optimization method and system, and the method comprises the steps: constructing a complete graph according to the obtained customer order data and an arc set between each pair of customer nodes, modeling a vehicle path problem VRPSPDT corresponding to the complete graph into a multi-target optimization problem, decomposing the multi-target optimization problem into a plurality of sub-problems, and carrying out the modeling of the sub-problems, and the plurality of sub-problems are iteratively solved for many times until a logistics goods taking and delivering path corresponding to the customer order data is obtained. In the nth iteration, operation of an evolutionary operator is carried out on current iteration information, a filial generation solution is obtained, the evolutionary operator comprises sequential crossover and random variation, the filial generation solution is input into a pre-trained deep reinforcement learning DRL model, the filial generation solution is optimized based on the DRL model, an improved solution is obtained, and the current population is updated based on the improved solution, and updating the current reference point according to the sub-target value of the filial generation solution. The efficiency, the accuracy and the reliability of determining the logistics goods taking and delivering path are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a method and system for optimizing logistics pick-up and delivery routes. Background Art

[0002] Green manufacturing and logistics play an important role in modern supply chain management. Manufacturing plants need to deliver goods purchased by customers online and collect discarded products from customers for reuse or proper disposal, which is called reverse logistics.

[0003] Due to its significant role in reducing energy consumption and environmental pollution, reverse logistics has been applied to distribution systems in various application scenarios, such as library book delivery, grocery delivery, and parcel delivery. In addition, in order to improve delivery efficiency, pickup and delivery services need to be completed within a predetermined time window. This type of actual logistics pickup and delivery problem is called the simultaneous pickup and delivery and vehicle routing problem with time window VRPSPDT.

[0004] Recent studies on solving VRPSPDT mainly use exact algorithms or heuristic algorithms, which are prone to falling into local optimality for two main reasons: 1) they usually only optimize a single weighted objective function during the entire search process, making it difficult to maintain the diversity of solutions; 2) they prioritize immediate improvements and ignore long-term rewards.

[0005] The content of the background technology section is only the information known to the inventor personally, and does not mean that the above information has entered the public domain before the application date of this disclosure, nor does it mean that it can become the prior art of the present disclosure. Summary of the invention

[0006] The present disclosure provides a logistics pick-up and delivery route optimization method and system to solve at least one of the above-mentioned technical problems.

[0007] In a first aspect, the present disclosure provides a method for optimizing a logistics pick-up and delivery route, comprising:

[0008] A complete graph is constructed based on the obtained customer order data and the arc set between each pair of customer nodes, and the vehicle routing problem VRPSPDT corresponding to the complete graph is modeled as a multi-objective optimization problem;

[0009] Decomposing the multi-objective optimization problem into multiple sub-problems, and solving the multiple sub-problems multiple times by iteration until a logistics pick-up and delivery path corresponding to the customer order data is obtained; wherein, in the nth iteration, n is an integer greater than 0, including:

[0010] Performing an evolutionary operator operation on the current iteration information to obtain a child solution, wherein the evolutionary operator includes sequential crossover and random mutation, and the current iteration information includes a current population and a current reference point formed by the minimum value of each target in the current population;

[0011] Inputting the child solution into a pre-trained deep reinforcement learning (DRL) model to optimize the child solution based on the DRL model to obtain an improved solution;

[0012] The current population is updated based on the improved solution, and the current reference point is updated according to the sub-objective value of the child solution.

[0013] In some embodiments, the sequential crossover is used to indicate that the child solution is determined by recombination and inheritance of the parent solution;

[0014] The random variation is used to indicate that the offspring solution is determined by random adjustment.

[0015] In some embodiments, the sequential interleaving comprises:

[0016] Randomly select individuals p from the first percentage of individuals in the current population 1 , randomly select individual p from the second percentage of individuals 2 , the sum of the first percentage and the second percentage is 100 percent;

[0017] Randomly generate two ascending integers pos in the interval [1,D] 1 and pos 2 , D is the total number of client nodes;

[0018] The p 1 POS 1 and pos 2 The sequence between is copied to the corresponding position of the sub-generation solution;

[0019] Use p 2 The unused elements in the offspring solution are sequentially filled to fill the remaining gene positions in the offspring solution to obtain a complete offspring individual.

[0020] In some embodiments, the random variation comprises:

[0021] Generate a random integer in [0,2];

[0022] If the random integer is 0, a random flip mutation operator is executed on the complete offspring individual to obtain the offspring solution;

[0023] If the random integer is 1, a random exchange mutation operator is performed on the complete offspring individual to obtain the offspring solution;

[0024] If the random integer is 2, a random repositioning mutation operator is executed on the complete offspring individual to obtain the offspring solution.

[0025] In some embodiments, the step of inputting the offspring solution into a pre-trained deep reinforcement learning (DRL) model to optimize the offspring solution based on the DRL model to obtain an improved solution comprises multiple iterative optimizations, wherein in the mth iteration, m is an integer greater than 0, including:

[0026] Obtaining a current state and a current action, and determining a current improved solution and a current reward value according to the current state corresponding to when the current action is executed;

[0027] If the current reward value is 0, the current unimproved flag is incremented by 1, otherwise the current unimproved flag is 0, and the current state is updated according to the current improved solution;

[0028] If the current unimproved flag is greater than the preset maximum number of unimproved steps, the iteration ends and the current improved solution is used as the improved solution.

[0029] In some embodiments, the DRL model is obtained based on experience replay training, and the experience replay includes:

[0030] The full state transition tuples obtained in each training iteration are stored in a preset playback pool, where each state transition tuple represents the state before and after the corresponding iteration, as well as the action, reward value, and unimproved flag of the corresponding iteration;

[0031] Randomly extracting some state transition tuples from the preset revisit pool;

[0032] A gradient descent operation is performed based on the partial state transfer tuple and a preset loss function to obtain the DRL model.

[0033] In some embodiments, the customer order data includes: delivery requirements, pickup requirements, start time of the time window, end time of the time window, and service time;

[0034] Among them, the distribution demand includes the quantity of goods transported from the warehouse to the customer node, the pick-up demand includes the quantity of goods retrieved by the customer node and returned to the warehouse, the start time includes the earliest time when the customer node accepts the pick-up and delivery service, the end time includes the latest time when the customer node accepts the pick-up and delivery service, and the service time includes the time consumed by the vehicle to load and unload the goods at the customer node.

[0035] In a second aspect, the present disclosure provides a logistics pick-up and delivery route optimization system, comprising:

[0036] A construction unit, used for constructing a complete graph according to the obtained customer order data and the arc set between each pair of customer nodes;

[0037] A modeling unit, used for modeling the vehicle routing problem VRPSPDT corresponding to the complete graph as a multi-objective optimization problem;

[0038] A decomposition unit, used for decomposing the multi-objective optimization problem into multiple sub-problems;

[0039] A solving unit, used for performing multiple iterations to solve the multiple sub-problems until a logistics pick-up and delivery path corresponding to the customer order data is obtained; wherein, in the nth iteration, n is an integer greater than 0, including:

[0040] Performing an evolutionary operator operation on the current iteration information to obtain a child solution, wherein the evolutionary operator includes sequential crossover and random mutation, and the current iteration information includes a current population and a current reference point formed by the minimum value of each target in the current population;

[0041] Inputting the child solution into a pre-trained deep reinforcement learning (DRL) model to optimize the child solution based on the DRL model to obtain an improved solution;

[0042] The current population is updated based on the improved solution, and the current reference point is updated according to the sub-objective value of the child solution.

[0043] In some embodiments, the sequential crossover is used to indicate that the child solution is determined by recombination and inheritance of the parent solution;

[0044] The random variation is used to indicate that the offspring solution is determined by random adjustment.

[0045] In some embodiments, the sequential interleaving comprises:

[0046] Randomly select individuals p from the first percentage of individuals in the current population 1 , randomly select individual p from the second percentage of individuals 2 , the sum of the first percentage and the second percentage is 100 percent;

[0047] Randomly generate two ascending integers pos in the interval [1,D] 1 and pos 2 , D is the total number of client nodes;

[0048] The p 1 POS 1 and pos 2 The sequence between is copied to the corresponding position of the sub-generation solution;

[0049] Use p2 The unused elements in the offspring solution are sequentially filled to fill the remaining gene positions in the offspring solution to obtain a complete offspring individual.

[0050] In some embodiments, the random variation comprises:

[0051] Generate a random integer in [0,2];

[0052] If the random integer is 0, a random flip mutation operator is executed on the complete offspring individual to obtain the offspring solution;

[0053] If the random integer is 1, a random exchange mutation operator is performed on the complete offspring individual to obtain the offspring solution;

[0054] If the random integer is 2, a random repositioning mutation operator is executed on the complete offspring individual to obtain the offspring solution.

[0055] In some embodiments, the solving unit optimizes the child solution based on the DRL model to obtain an improved solution, including multiple iterative optimizations, in the mth iteration, where m is an integer greater than 0, including:

[0056] Obtaining a current state and a current action, and determining a current improved solution and a current reward value according to the current state corresponding to when the current action is executed;

[0057] If the current reward value is 0, the current unimproved flag is incremented by 1, otherwise the current unimproved flag is 0, and the current state is updated according to the current improved solution;

[0058] If the current unimproved flag is greater than the preset maximum number of unimproved steps, the iteration ends and the current improved solution is used as the improved solution.

[0059] In some embodiments, the DRL model is obtained based on experience replay training, and the experience replay includes:

[0060] The full state transition tuples obtained in each training iteration are stored in a preset playback pool, where each state transition tuple represents the state before and after the corresponding iteration, as well as the action, reward value, and unimproved flag of the corresponding iteration;

[0061] Randomly extracting some state transition tuples from the preset revisit pool;

[0062] A gradient descent operation is performed based on the partial state transfer tuple and a preset loss function to obtain the DRL model.

[0063] In some embodiments, the customer order data includes: delivery requirements, pickup requirements, start time of the time window, end time of the time window, and service time;

[0064] Among them, the distribution demand includes the quantity of goods transported from the warehouse to the customer node, the pick-up demand includes the quantity of goods retrieved by the customer node and returned to the warehouse, the start time includes the earliest time when the customer node accepts the pick-up and delivery service, the end time includes the latest time when the customer node accepts the pick-up and delivery service, and the service time includes the time consumed by the vehicle to load and unload the goods at the customer node.

[0065] In a third aspect, the present disclosure provides a processor-readable storage medium, wherein the processor-readable storage medium stores a computer program, wherein the computer program is used to enable the processor to execute the method described in the first aspect above.

[0066] In a fourth aspect, the present disclosure provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0067] The logistics pick-up and delivery path optimization method and system provided by the present disclosure include: constructing a complete graph according to the obtained customer order data and the arc set between each pair of customer nodes, and modeling the vehicle path problem VRPSPDT corresponding to the complete graph as a multi-objective optimization problem, decomposing the multi-objective optimization problem into multiple sub-problems, and solving the multiple sub-problems for multiple iterations until the logistics pick-up and delivery path corresponding to the customer order data is obtained. Wherein, in the nth iteration, n is an integer greater than 0, including: performing an evolutionary operator operation on the current iteration information to obtain a sub-generation solution, wherein the evolutionary operator includes sequential crossover and random mutation, the current iteration information includes the current population and the current reference point composed of the minimum value of each target in the current population, inputting the sub-generation solution into a pre-trained deep reinforcement learning DRL model, optimizing the sub-generation solution based on the DRL model, obtaining an improved solution, updating the current population based on the improved solution, and updating the current reference point according to the sub-target value of the sub-generation solution. On the one hand, the diversity of solutions is enhanced by sequential crossover and random mutation in the evolutionary algorithm, providing powerful exploration capabilities and diverse experience for solving actual VRPSPDT. On the other hand, by developing high-quality solutions to solve the actual VRPSPDT through the DRL model, learning from network-specific knowledge and past experience, the DRL model can effectively enhance the development capability, thereby accelerating the population convergence. On the other hand, VRPSPDT is first transformed into a multi-objective optimization problem (MOP) through a multi-objective strategy, and then the MOP is decomposed into multiple sub-problems. The evolutionary operation based on the evolutionary algorithm and the DRL model are used to improve the diversity and quality of the solution, thereby providing a relatively better overall performance for solving the actual VRPSPDT. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0069] Figure 1 A flow chart of the logistics pick-up and delivery route optimization method provided by the present disclosure;

[0070] Figure 2 A schematic diagram of a scenario for the logistics pick-up and delivery route optimization method provided in the present disclosure;

[0071] Figure 3 A schematic diagram of the principle of the logistics pick-up and delivery route optimization method provided by the present disclosure;

[0072] Figure 4 A schematic diagram of the principle of sequential crossover provided for the present disclosure;

[0073] Figure 5 A schematic diagram of the principle of random mutation provided by the present disclosure;

[0074] Figure 6 Schematic diagram of the principles of five local search strategies provided in the present disclosure;

[0075] Figure 7 A schematic diagram of the principle of the training phase of the DRL model provided in the present disclosure;

[0076] Figure 8 A schematic diagram of the framework of the logistics pick-up and delivery route optimization system provided by the present disclosure;

[0077] Fig. 9 A schematic diagram of the framework of an electronic device provided by the present disclosure.

[0078] The above drawings show clear embodiments of the present disclosure, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present disclosure in any way, but to illustrate the concepts of the present disclosure to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0079] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0080] It should be understood that the terms "including" and "having" and any variations thereof in the embodiments of the present disclosure are intended to cover but not exclude inclusion. For example, a product or device comprising a series of components is not necessarily limited to those components explicitly listed, but may include other components not explicitly listed or inherent to these products or devices.

[0081] In the embodiments of the present disclosure, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0082] The term "plurality" in the embodiments of the present disclosure refers to two or more than two, and other quantifiers are similar thereto.

[0083] The terms "first", "second", "third", etc. in the present disclosure are used to distinguish similar or similar objects or entities, and are not necessarily meant to limit a specific order or sequence, unless otherwise indicated. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, for example, they can be implemented in an order other than those given in the diagrams or descriptions of the embodiments of the present disclosure.

[0084] The term "unit / module" as used in this disclosure refers to any known or later developed hardware, software, firmware, artificial intelligence, fuzzy logic or a combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0085] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0086] According to one aspect of the present disclosure, the present disclosure provides a method for optimizing logistics pick-up and delivery routes.

[0087] See also Figure 1 , Figure 1 The following is a flow chart of the logistics pick-up and delivery route optimization method provided by the present disclosure. Figure 1 As shown, the method includes:

[0088] S101: Based on the obtained customer order data and the arc set between each pair of customer nodes, a complete graph is constructed, and the vehicle routing problem VRPSPDT corresponding to the complete graph is modeled as a multi-objective optimization problem.

[0089] The execution subject of the logistics pick-up and delivery path optimization method provided in the present disclosure is a logistics pick-up and delivery path optimization system. The logistics pick-up and delivery path optimization system can be a server, a terminal device, a processor, etc., which are not listed here one by one. The description of the logistics pick-up and delivery path optimization system is explained later.

[0090] The logistics delivery route optimization method provided in this manual is applicable to: Figure 2 The logistics scenario shown in Figure 2. Figure 2 As shown in the figure, the logistics scenario includes multiple vehicles, warehouses, and multiple customer nodes. Different customer nodes correspond to the delivery or pickup time periods. The goal of VRPSPDT is to plan the minimum cost route for multiple vehicles while meeting various constraints, and provide pickup and delivery services for many customer nodes within the specified time window.

[0091] After receiving the customer order data V requested by the customer node and the arc set E between each customer node, the logistics delivery route optimization system can perform object processing to obtain a complete graph G = (V, E). Among them, V = {0, 1, 2, ..., M} contains warehouse 0 and M customer nodes. E = {<i,j> |i, j∈V, i≠j} represents the set of arcs between each pair of client nodes.

[0092] The solution of VRPSPDT consists of a set of vehicle routes, denoted as sol = {R1, R2, ..., RK}, where K is the number of vehicles used in sol. Each route Ri represents a series of customer nodes visited by the vehicle, denoted as Ri = (hi, 1, hi, 2, ..., hi, Li), where hi, j represents the jth customer node visited on route Ri, and Li represents the length of route Ri, i.e., the number of customer nodes served. The goal of VRPSPDT is to minimize the total cost (TC), which can be expressed by Equation 1:

[0093]

[0094] Among them, f1 is the vehicle scheduling cost, and f 1 =u 1 K,u 1 is the dispatch cost of each vehicle, f2 is the transportation cost, and f 2 =u 2 TD, and TD is the total travel distance, u 2 is the transportation cost per unit distance.

[0095] Therefore, the logistics delivery path optimization system can model VRPSPDT as a system that contains two sub-goals (f 1 and f 2) multi-objective optimization problem.

[0096] In some embodiments, customer order data includes: delivery requirements d i 、Pick-up demand i , the start time of the time window a i , the end time of the time window b i , Service time i .

[0097] Among them, the distribution demand d i Including the quantity of goods transported from the warehouse to customer node i, the pick-up demand p i Includes the number of goods retrieved by customer node i and returned to the warehouse, starting time a i Including the earliest time when customer node i accepts the delivery service, the termination time b i Including the latest time that customer node i accepts the delivery service, service time s i Includes the time consumed by the vehicle to load and unload goods at customer node i.

[0098] Correspondingly, client node i can be represented by (d i ,p i ,a i ,b i ,s i )express.

[0099] S102: Decompose the multi-objective optimization problem into multiple sub-problems, and perform multiple iterations to solve the multiple sub-problems until a logistics pick-up and delivery path corresponding to the customer order data is obtained. In the nth iteration, n is an integer greater than 0, including:

[0100] Step 1: Perform an evolutionary operator operation on the current iteration information to obtain a sub-generation solution, wherein the evolutionary operator includes sequential crossover and random mutation, and the current iteration information includes the current population and the current reference point consisting of the minimum value of each target in the current population.

[0101] Step 2: Input the offspring solution into a pre-trained deep reinforcement learning (DRL) model to optimize the offspring solution based on the DRL model to obtain an improved solution.

[0102] Step 3: Update the current population based on the improved solution, and update the current reference point according to the sub-objective value of the offspring solution.

[0103] Combined with the above analysis, VRPSPDT is modeled as a logistics delivery path optimization system with two sub-goals (f 1 and f 2), the multi-objective optimization problem can be further decomposed into a series of sub-problems.

[0104] These sub-problems are solved by combining the evolutionary operation based on the evolutionary algorithm and the DRL operation of the DRL model. Relatively speaking, the evolutionary operation provides the DRL model with diverse experience through population evolution, while the DRL model uses all the information and network knowledge of the Deep Q-Network (DQN) to accelerate the convergence of the population in the evolutionary operation. In order to select the most optimal solution from the population, the VRPSPDT task is effectively completed, that is, the logistics pick-up and delivery path that satisfies the minimum total cost (TC) is obtained.

[0105] That is to say, combined Figure 3 It can be seen that in this embodiment, the logistics pick-up and delivery path optimization system determines the logistics pick-up and delivery path that satisfies the minimum total cost (TC) by combining the evolutionary operation and the DRL operation of the DRL model.

[0106] Combination Figure 3 It can be seen that in the evolutionary operation, the operation of initializing the population can be performed to obtain the weight vectors of the T nearest Euclidean distances generated by each subproblem, the initialization population (generate N random sequences and use the split method to generate N solutions), and the reference point composed of the minimum value of each target in the initial population. T and N are both integers greater than 0).

[0107] After the population is initialized, the evolution operator can be operated. Figure 3 As shown, the evolutionary operators include sequential crossover and random mutation, aiming to enhance the exploration ability.

[0108] In some embodiments, sequential interleaving includes:

[0109] Step 1: Randomly select individuals p from the first percent of individuals in the current population 1 , randomly select individual p from the second percentage of individuals 2 , the sum of the first percentage and the second percentage is one hundred percent.

[0110] For example, the first percentage and the second percentage are 50% respectively.

[0111] Step 2: Randomly generate two ascending integers pos in the interval [1, D] 1 and pos 2 , D is the total number of client nodes.

[0112] That is to say, two ascending integers are randomly generated within the range of the number of all client nodes.

[0113] Step 3: Place p 1POS 1 and pos 2 The sequence between them is copied to the corresponding position of the child solution.

[0114] Step 4: Use p 2 The unused elements in the are used to fill the remaining gene positions in the offspring solution in order to obtain complete offspring individuals.

[0115] like Figure 4 As shown, sequential crossover explores high-quality child solutions by recombining and inheriting parent solutions. For example, the two parent solutions are divided into parent solution 1 and parent solution 2. Parent solution 1 is customer node 2->4->7->1->8->3->6->5. Parent solution 2 is customer node 8->1->2->5->6->4->7->3. First, a part of the fragments is randomly selected from parent solution 1, where "customer node->1->8->3" is selected. Then, this part of the fragment is copied to the child solution. Next, genes (ie, numbers) that do not appear in the child solution are selected from parent solution 2, and they are filled into the remaining positions of the child solution in the order in which they appear in parent solution 2. The child solution finally generated is customer node 2->5->7->1->8->3->6->4.

[0116] In some embodiments, random variation comprises:

[0117] Step 1: Generate a random integer in [0,2].

[0118] Exemplarily, when the sequential crossover reaches a preset condition, random mutation is initiated. For example, the preset condition may be that the sequential crossover reaches a preset mutation probability.

[0119] Step 2: If the random integer is 0, a random flip mutation operator is executed on the complete offspring individual to obtain an offspring solution.

[0120] Step 3: If the random integer is 1, the random exchange mutation operator is executed on the complete offspring individuals to obtain the offspring solution.

[0121] Step 4: If the random integer is 2, the random relocation mutation operator is executed on the complete offspring individuals to obtain the offspring solution.

[0122] like Figure 5 As shown in Figure 1, random mutation includes random inversion, random swap, and random relocation. Random mutation explores new solutions with good genetic diversity by randomly adjusting the offspring solutions.

[0123] For example, in random flipping, you can select a continuous segment in the parent solution and flip the order of the elements in this segment. Figure 5As shown, the segment of customer nodes 3->2->1 is selected and flipped to customer nodes 1->2->3.

[0124] In random swapping, two different client nodes are selected in the parent solution. These two client nodes are swapped. Figure 5 As shown, customer node 4 and customer node 5 are selected, and customer nodes 4 and 5 are exchanged.

[0125] In random relocation, a client node is selected in the parent solution. This client node is moved to another randomly selected location. Figure 5 As shown, customer node 5 is selected and moved from customer node 5 to between customer node 2 and customer node 1.

[0126] Continue reading Figure 3 After random mutation, the optimized offspring solution is obtained through DRL operation, that is, the improved solution. In the evolutionary operation, the population can be updated according to the improved solution, and the reference point can be updated accordingly according to the size of the sub-objective value of the improved solution.

[0127] Then, it can be determined whether this update meets the termination condition. If so, the solution that satisfies the minimum of Formula 1 can be selected from the population to obtain the logistics pick-up and delivery path; otherwise, if not, the operation of the evolutionary operator is performed, and so on, until the iterative update meets the termination condition.

[0128] In some embodiments, the above step 2 includes multiple iterative optimizations, that is, the DRL operation is a process of multiple iterative optimizations. In the mth iteration, m is an integer greater than 0, including:

[0129] Step 21: Obtain the current state and the current action, and determine the current improved solution and the current reward value according to the current state corresponding to the execution of the current action.

[0130] Step 22: If the current reward value is 0, the current unimproved flag is incremented by 1, otherwise the current unimproved flag is 0, and the current state is updated according to the current improved solution.

[0131] Step 23: If the current unimproved flag is greater than the preset maximum number of unimproved steps, the iteration ends and the current improved solution is used as the improved solution.

[0132] Exemplarily, the DRL operation includes two inputs: the child solution and the preset maximum number of steps without improvement NO_IMP. First, the DRL operation deletes the warehouse delimiters in all paths of the child solution and concatenates the remaining sequences to obtain the current state s of the DRL operation.

[0133] Next, the DRL operation enters the iterative process: in the mth iteration, an action a is selected according to the greedy strategy. Specifically, the current action a can be obtained randomly according to the preset probability ε, or a = argmax a , Q(s,a′;θ) is used to determine the current action a.

[0134] Among them, θ is a random weight, a′ is a candidate action (i.e. a possible action), argmax a , Q is the maximized action value function of all possible actions. The Q function (also called action value function) can be approximated using a network with two fully connected layers, and the Q function can be expressed by Formula 2:

[0135]

[0136] Among them, r t 、r t+1 、r t+2 , is the immediate reward, indicating the reward value obtained at time steps t, t+1, and t+2; γ is the discount factor; s t is the state at time step t; a t is the action at time t; π is the greedy strategy.

[0137] That is, the logistics delivery route optimization system uses the greedy strategy π at each time step t to obtain the maximum expected γ discounted reward sum by observing the current state s and taking the current action a. The current action a is applied to the offspring solution to obtain the current improved solution and the current reward value (TC value Δimp). The new state s′ is generated by encoding the updated solution. If the DRL operation fails to improve the solution within the maximum number of steps without improvement NO_IMP steps, it means that the current solution cannot be further improved and marks the end of the DRL operation in this round.

[0138] For example, in one iteration, a logistics pick-up and delivery routing optimization system can:

[0139] 1) Use the greedy strategy π to select a local search strategy as the action for this iteration;

[0140] 2) Execute the action of this iteration on the child solution of this iteration, and generate the improved solution of this iteration and the corresponding TC value Δimp;

[0141] 3) Set the reward value of this iteration to Δimp;

[0142] 4) If the reward value of this iteration is 0, the unimproved flag no_imp corresponding to this iteration is incremented by 1; otherwise, the unimproved flag no_imp corresponding to this iteration = 0;

[0143] 5) If the reward value of this iteration is greater than , there is no improvement in the maximum number of steps NO_IMP, and the completion flag done = 1;

[0144] 6) If done = 1, the current iteration process ends and the improved solution (i.e., the improved solution) is output.

[0145] In some embodiments, the greedy strategy is Figure 6 One of the five local search strategies in . Figure 6 As shown, the five local search strategies include:

[0146] The exchange local search strategy can be understood as selecting two client nodes and exchanging their positions. This can change the order of the path, thus potentially finding a shorter path. Figure 6 As shown, the initial paths are customer nodes 0->1->3->0 and customer nodes 0->2->4->0. By swapping the positions of customer nodes 3 and 4, the paths become customer nodes 0->1->4->0 and customer nodes 0->2->3->0.

[0147] Relocate local search strategy can be understood as selecting a client node and moving it to another location. This can reorganize the path to reduce the total distance. Figure 6 As shown, the initial path is customer nodes 0->1->3->0 and 0->2->4->0. By moving customer node 3 to other positions in the path, the path becomes customer nodes 0->1->4->0 and 0->2->3->0.

[0148] The Cross-Exchange local search strategy can be understood as selecting two client nodes and performing a cross-exchange. This means that not only the positions of the client nodes are exchanged, but also the path segments between them may be exchanged. Figure 6 As shown, the initial path is customer nodes 0->1->6->4->0 and 0->2->5->3->0. By cross-exchanging the positions of customer nodes 6 and 5 and their path segments, the path becomes customer nodes 0->1->5->4->0 and 0->2->6->3->0.

[0149] Or-Opt local search strategy can be understood as selecting a customer node and trying to insert it into different positions of the path to find the optimal solution. Figure 6 As shown, the initial paths are customer nodes 0->1->6->3->0 and 0->2->5->4->0. By inserting customer node 6 at different positions of the path, the paths become customer nodes 0->1->5->4->0 and 0->2->6->3->0.

[0150] 2-Optimization (2-Opt) local search strategy can be understood as selecting two edges, deleting them, and then reconnecting the two ends of the path to form a new path. This can significantly reduce the path length. Figure 6 As shown, the initial path is customer nodes 0->1->5->4->0 and 0->2->6->3->0. By deleting the edges (1,5) and (4,0) and reconnecting the two ends of the path, the path becomes customer nodes 0->1->6->3->0 and 0->2->5->4->0.

[0151] In some embodiments, the DRL model is obtained based on experience replay training, and the experience replay includes:

[0152] Step 1: Store all state transition tuples obtained in each training iteration into the preset replay pool. Each state transition tuple represents the state before and after the corresponding iteration, as well as the action, reward value, and unimproved flag of the corresponding iteration.

[0153] Step 2: Randomly extract some state transition tuples from the preset revisit pool.

[0154] Step 3: Perform a gradient descent operation based on the partial state transition tuple and the preset loss function to obtain the DRL model.

[0155] For example, combining the above examples and Figure 7 It can be seen that in the training phase, when obtaining the sub-solution of the training phase (such as Figure 7 After that, you can also delete the warehouse delimiter to get the current state of the training phase (such as Figure 7 Then, the new state can be determined based on the current state of the training phase (e.g. Figure 7 The specific implementation principle can be found in the above description of the DRL model in the inference phase, which will not be repeated here.

[0156] Correspondingly, during experience replay, the state transition tuple (s, a, r, s′, done) is stored in a preset replay pool (e.g. Figure 7 Then randomly sample small batches of state transition tuples from the experience replay pool for training. For each sampled state transition tuple, if done j The flag is equal to 1, then the target value γ j Set to the immediate reward value r. Otherwise, y j Add the immediate reward value through the target network Estimated maximum future discounted reward value Target Network In order to avoid instability during training, an additional network is introduced. Then, by minimizing the predicted value and the target value y j The loss between θ and θ is used to update the weight θ of the deep Q network using gradient descent. The loss function and the update formula of the deep Q network can be found in Equation 3 and Equation 4, where Equation 3:

[0157] L(θ)=[yQ(s,a;θ)] 2

[0158] Among them, L(θ) is the loss function, y is the target value, and Q(s,a;θ) is the predicted value.

[0159] Formula 4:

[0160]

[0161] Among them, θ t+1 is the weight at time step t+1, θ t is the weight of time step t.

[0162] In addition, the target network The weights of are updated to the weights of the deep Q network every N steps. Since the data collected in the DRL model are usually correlated, sequential training has a negative impact on the convergence and stability of the neural network. To alleviate this problem, the training of the DRL model uses an experience replay pool to destroy data correlation. At the same time, a separate target network is used to delay weight updates, which effectively solves the inherent instability problem in the DRL model.

[0163] That is, the experience replay during the training phase includes:

[0164] 1) Store the state transition tuple (s, a, r, s′, done) in the preset playback pool;

[0165] 2) Randomly extract a small batch of state transfer tuples (s) from the preset playback pool j , a j , r j , s′ j , done j );

[0166] 3) If done j If the flag is equal to 1, the target value y j is the immediate reward value r. Otherwise, y j for

[0167] 4) Perform a gradient descent operation on θ in equation 3;

[0168] 5) In the N-step cycle, reset That is, θ - =θ.

[0169] According to one aspect of the present disclosure, the present disclosure provides a logistics pick-up and delivery route optimization system.

[0170] See also Figure 8 , Figure 8 This is a schematic diagram of the framework of the logistics delivery route optimization system provided by the present disclosure. Figure 8 As shown, the system 800 includes:

[0171] A construction unit 801 is used to construct a complete graph according to the obtained customer order data and the arc set between each pair of customer nodes;

[0172] A modeling unit 802, configured to model the vehicle routing problem VRPSPDT corresponding to the complete graph as a multi-objective optimization problem;

[0173] A decomposition unit 803 is used to decompose the multi-objective optimization problem into multiple sub-problems;

[0174] The solving unit 804 is used to perform multiple iterations to solve the multiple sub-problems until a logistics pick-up and delivery path corresponding to the customer order data is obtained; wherein, in the nth iteration, n is an integer greater than 0, including:

[0175] Performing an evolutionary operator operation on the current iteration information to obtain a child solution, wherein the evolutionary operator includes sequential crossover and random mutation, and the current iteration information includes a current population and a current reference point formed by the minimum value of each target in the current population;

[0176] Inputting the child solution into a pre-trained deep reinforcement learning (DRL) model to optimize the child solution based on the DRL model to obtain an improved solution;

[0177] The current population is updated based on the improved solution, and the current reference point is updated according to the sub-objective value of the child solution.

[0178] In some embodiments, the sequential crossover is used to indicate that the child solution is determined by recombination and inheritance of the parent solution;

[0179] The random variation is used to indicate that the offspring solution is determined by random adjustment.

[0180] In some embodiments, the sequential interleaving comprises:

[0181] Randomly select individuals p from the first percentage of individuals in the current population 1 , randomly select individual p from the second percentage of individuals 2, the sum of the first percentage and the second percentage is 100 percent;

[0182] Randomly generate two ascending integers pos in the interval [1,D] 1 and pos 2 , D is the total number of client nodes;

[0183] The p 1 POS 1 and pos 2 The sequence between is copied to the corresponding position of the sub-generation solution;

[0184] Use p 2 The unused elements in the offspring solution are sequentially filled to fill the remaining gene positions in the offspring solution to obtain a complete offspring individual.

[0185] In some embodiments, the random variation comprises:

[0186] Generate a random integer in [0,2];

[0187] If the random integer is 0, a random flip mutation operator is executed on the complete offspring individual to obtain the offspring solution;

[0188] If the random integer is 1, a random exchange mutation operator is performed on the complete offspring individual to obtain the offspring solution;

[0189] If the random integer is 2, a random repositioning mutation operator is executed on the complete offspring individual to obtain the offspring solution.

[0190] In some embodiments, the solving unit 804 optimizes the child solution based on the DRL model to obtain an improved solution, including multiple iterative optimizations, in the mth iteration, where m is an integer greater than 0, including:

[0191] Obtaining a current state and a current action, and determining a current improved solution and a current reward value according to the current state corresponding to when the current action is executed;

[0192] If the current reward value is 0, the current unimproved flag is incremented by 1, otherwise the current unimproved flag is 0, and the current state is updated according to the current improved solution;

[0193] If the current unimproved flag is greater than the preset maximum number of unimproved steps, the iteration ends and the current improved solution is used as the improved solution.

[0194] In some embodiments, the DRL model is obtained based on experience replay training, and the experience replay includes:

[0195] The full state transition tuples obtained in each training iteration are stored in a preset playback pool, where each state transition tuple represents the state before and after the corresponding iteration, as well as the action, reward value, and unimproved flag of the corresponding iteration;

[0196] Randomly extracting some state transition tuples from the preset revisit pool;

[0197] A gradient descent operation is performed based on the partial state transfer tuple and a preset loss function to obtain the DRL model.

[0198] In some embodiments, the customer order data includes: delivery requirements, pickup requirements, start time of the time window, end time of the time window, and service time;

[0199] Among them, the distribution demand includes the quantity of goods transported from the warehouse to the customer node, the pick-up demand includes the quantity of goods retrieved by the customer node and returned to the warehouse, the start time includes the earliest time when the customer node accepts the pick-up and delivery service, the end time includes the latest time when the customer node accepts the pick-up and delivery service, and the service time includes the time consumed by the vehicle to load and unload the goods at the customer node.

[0200] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0201] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0202] Fig. 9 A schematic block diagram of an example electronic device 900 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0203] like Fig. 9As shown, the device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0204] A number of components in the device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0205] The computing unit 901 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods and processes described above, such as a logistics pickup and delivery path optimization method. For example, in some embodiments, the logistics pickup and delivery path optimization method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the logistics pickup and delivery path optimization method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform the logistics pickup and delivery path optimization method in any other appropriate manner (e.g., by means of firmware).

[0206] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0207] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0208] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0209] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0210] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0211] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0212] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0213] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer executable instructions. These computer executable instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0214] These processor executable instructions may also be stored in a processor readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the processor readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0215] These processor-executable instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0216] Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is also intended to include these modifications and variations.

Claims

1. A method for optimizing logistics pick-up and delivery routes, characterized in that: The method comprises: A complete graph is constructed based on the obtained customer order data and the arc set between each pair of customer nodes, and the vehicle routing problem VRPSPDT corresponding to the complete graph is modeled as a multi-objective optimization problem; Decomposing the multi-objective optimization problem into multiple sub-problems, and solving the multiple sub-problems multiple times by iteration until a logistics pick-up and delivery path corresponding to the customer order data is obtained; wherein, in the nth iteration, n is an integer greater than 0, including: Performing an evolutionary operator operation on the current iteration information to obtain a child solution, wherein the evolutionary operator includes sequential crossover and random mutation, and the current iteration information includes a current population and a current reference point formed by the minimum value of each target in the current population; Inputting the child solution into a pre-trained deep reinforcement learning (DRL) model to optimize the child solution based on the DRL model to obtain an improved solution; The current population is updated based on the improved solution, and the current reference point is updated according to the sub-objective value of the child solution.

2. The method according to claim 1, characterized in that The sequential crossover is used to indicate that the child solution is determined by recombining and inheriting the parent solution; The random variation is used to indicate that the offspring solution is determined by random adjustment.

3. The method according to claim 1, characterized in that The sequential crossover includes: Randomly select an individual p1 from the first percentage of individuals in the current population, and randomly select an individual p2 from the second percentage of individuals, wherein the sum of the first percentage and the second percentage is 100 percent; Randomly generate two ascending integers pos1 and pos2 in the interval [1, D], where D is the total number of client nodes; Copy the sequence between pos1 and pos2 in p1 to the corresponding position of the child solution; The remaining gene positions in the offspring solution are filled in order with the unused elements in p2 to obtain complete offspring individuals.

4. The method according to claim 3, characterized in that The random variation includes: Generate a random integer in [0,2]; If the random integer is 0, a random flip mutation operator is executed on the complete offspring individual to obtain the offspring solution; If the random integer is 1, a random exchange mutation operator is performed on the complete offspring individual to obtain the offspring solution; If the random integer is 2, a random repositioning mutation operator is executed on the complete offspring individual to obtain the offspring solution.

5. The method according to claim 3, characterized in that: The step of inputting the offspring solution into a pre-trained deep reinforcement learning (DRL) model to optimize the offspring solution based on the DRL model to obtain an improved solution comprises multiple iterative optimizations, wherein in the mth iteration, m is an integer greater than 0, including: Obtaining a current state and a current action, and determining a current improved solution and a current reward value according to the current state corresponding to when the current action is executed; If the current reward value is 0, the current unimproved flag is incremented by 1, otherwise the current unimproved flag is 0, and the current state is updated according to the current improved solution; If the current unimproved flag is greater than the preset maximum number of unimproved steps, the iteration ends and the current improved solution is used as the improved solution.

6. According to the method of claim 1, the DRL model is obtained based on experience replay training, and the experience replay includes: The full state transition tuples obtained in each training iteration are stored in a preset playback pool, where each state transition tuple represents the state before and after the corresponding iteration, as well as the action, reward value, and unimproved flag of the corresponding iteration; Randomly extracting some state transition tuples from the preset revisit pool; A gradient descent operation is performed based on the partial state transfer tuple and a preset loss function to obtain the DRL model.

7. The method according to claim 1, wherein the customer order data comprises: Delivery requirements, pickup requirements, start time of the time window, end time of the time window, service time; Among them, the distribution demand includes the quantity of goods transported from the warehouse to the customer node, the pick-up demand includes the quantity of goods retrieved by the customer node and returned to the warehouse, the start time includes the earliest time when the customer node accepts the pick-up and delivery service, the end time includes the latest time when the customer node accepts the pick-up and delivery service, and the service time includes the time consumed by the vehicle to load and unload the goods at the customer node.

8. A logistics delivery route optimization system, characterized in that: include: A construction unit, used for constructing a complete graph according to the obtained customer order data and the arc set between each pair of customer nodes; A modeling unit, used for modeling the vehicle routing problem VRPSPDT corresponding to the complete graph as a multi-objective optimization problem; A decomposition unit, used for decomposing the multi-objective optimization problem into multiple sub-problems; A solving unit, used for performing multiple iterations to solve the multiple sub-problems until a logistics pick-up and delivery path corresponding to the customer order data is obtained; wherein, in the nth iteration, n is an integer greater than 0, including: Performing an evolutionary operator operation on the current iteration information to obtain a child solution, wherein the evolutionary operator includes sequential crossover and random mutation, and the current iteration information includes a current population and a current reference point formed by the minimum value of each target in the current population; Inputting the child solution into a pre-trained deep reinforcement learning (DRL) model to optimize the child solution based on the DRL model to obtain an improved solution; The current population is updated based on the improved solution, and the current reference point is updated according to the sub-objective value of the child solution.

9. A processor-readable storage medium, characterized in that: The processor-readable storage medium stores a computer program, and the computer program is used to cause the processor to execute the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Logistics distribution path acquisition method, apparatus and device, and storage medium

    CN122022659A

  • A logistics distribution path acquisition method, device, equipment and storage medium

    CN122022659B