Multi-warehouse distribution task path planning method and system

Abstract

The invention discloses a path planning method and system for a multi-warehouse distribution task, and relates to the field of path planning, and the method comprises the steps: constructing a vehicle distribution task model which comprises three objective functions and a plurality of constraint conditions as an upper layer model, and a lower layer model for optimizing the objective functions through employing a simulated annealing method; generating a new path planning scheme by using a genetic algorithm according to the minimum travel cost, the sum of the maximum warehouse stock and the path planning scheme corresponding to the maximum robustness; and a vehicle distribution task model is used for the new path planning scheme, solving is carried out through a non-dominated sorting genetic algorithm, and a path plan of a distribution task is obtained. According to the method, the optimal planning path capable of adapting to various scenes can be obtained.

Description

Technical Field

[0001] The present invention relates to the field of path planning, and in particular to a path planning method and system for multi-warehouse distribution tasks. Background Art

[0002] The Vehicle Routing Problem (VRP) is a fundamental problem in operations research and logistics that focuses on optimizing the routing of vehicles to efficiently serve a set of customers. It consists of finding the best possible routes for a fleet of vehicles subject to various constraints, such as vehicle capacities, travel times, and service preferences. Traditional VRPs, such as Capacitated Vehicle Routing Problem (CVRP) and Vehicle Routing Problem with Time Windows (VRPTWs), usually assume that all vehicles start and end their routes at a single, well-stocked depot, serving only a set of customers with predefined and fixed demands.

[0003] In the prior art, the planning of multi-warehouse vehicle routing problems adopts the overall method, two-stage method and two-layer optimization. The overall method usually establishes a virtual center to transform the multi-warehouse problem into a whole; the two-stage method first groups customers and then decomposes them into multiple single-warehouse routing planning problems for solution; the two-layer optimization method establishes upper and lower layer related problem models for customer allocation and routing optimization of such problems, alternately iterates the overall routing planning, and can conduct a comprehensive search in the decision space to obtain a satisfactory delivery plan. However, such methods usually consume a lot of computing resources, and at the same time, they need to sample and evaluate multiple scenarios under uncertain requirements, which further aggravates the consumption of computing resources.

[0004] In summary, the above technologies cannot obtain the optimal planning path that can adapt to various scenarios. Summary of the invention

[0005] The embodiments of the present invention provide a path planning method and system for multi-warehouse distribution tasks, which can solve the problem in the prior art that an optimal planning path that can adapt to a variety of scenarios cannot be obtained.

[0006] The embodiment of the present invention provides a path planning method for a multi-warehouse delivery task, the path planning method being applied to a vehicle delivery task consisting of M warehouses, N task nodes with delivery time constraints, and K indifferent vehicles with capacity constraints; the path planning method comprises the following steps:

[0007] The travel cost reflecting the cost of the vehicle's driving path, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model are used as objective functions; each customer is assigned to only one customer, each customer receives only one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive the delivery service of the same vehicle, the initial location and the end location of each vehicle's delivery route must be in the same warehouse, and sub-routes must be eliminated as constraints, so as to construct a vehicle distribution task model that can generate multiple path planning solutions and use the vehicle distribution task model as the upper model of the two-layer optimization model;

[0008] The minimum travel cost is selected from the travel costs of multiple path planning schemes randomly generated by the upper model through simulated annealing, and the path planning scheme corresponding to the minimum travel cost is obtained to construct the lower model of the two-layer optimization model;

[0009] The multiple path planning schemes obtained from the two-layer optimization model are used as the initial population of the simulated annealing genetic algorithm to train the Kriging model, and the path distance fitting model used to describe the correlation between customer allocation relationship and path distance is obtained, and the path distance fitting model is used as the proxy model.

[0010] A plurality of path planning schemes as the initial population are subjected to crossover mutation to generate offspring, and the offspring are input into the proxy model for evaluation; based on the evaluation results, a solution that is not dominated by any other solution is found from all solutions using non-dominated sorting, and a new offspring is obtained through a local search method and a simulated annealing method; the proxy model is retrained using the new offspring to update the proxy model and the parent generation is updated through an environmental selection strategy; the simulated annealing genetic algorithm is stopped when the termination condition is met, and the new offspring is used as the optimal planning path.

[0011] Furthermore, the step of selecting the minimum travel cost from the travel costs of multiple path planning solutions randomly generated by the upper model by simulated annealing includes:

[0012] Randomly generate multiple path planning solutions that meet the objective function and constraints;

[0013] Obtaining solution results of the first objective function, the second objective function and the third objective function of the initial population;

[0014] Designing a simulated annealing method, including: setting the initial temperature to 1000 degrees Celsius, the stop temperature to 1 degree Celsius, and the annealing rate to 0.999 degrees Celsius / second;

[0015] Using a simulated annealing method to process the solution results of the first objective function, the second objective function and the third objective function to obtain a final solution;

[0016] In the iterative process of the simulated annealing, a local search heuristic algorithm optimized by reinsertion, Swap or 2-Opt is used to generate offspring, and a value suitable for the objective function is selected from the offspring and the final solution to update the final solution.

[0017] Furthermore, the step of obtaining a path distance fitting model for describing the correlation between the customer allocation relationship and the path distance, and using the path distance fitting model as a proxy model, specifically includes the following steps:

[0018] Extracting two characteristic parameters of each solution according to the multiple path planning solutions as the initial population; wherein the two characteristic parameters are the average distance from the customer to the warehouse and the aggregation degree of the customer allocation solution;

[0019] The Kriging model is trained based on the two characteristic parameters to obtain the proxy model.

[0020] Furthermore, the steps of performing crossover mutation on the multiple path planning schemes as the initial population to generate offspring include:

[0021] Based on multiple path planning schemes as the initial population, a binary bidding method is used to generate a The mating pool contains the path planning scheme corresponding to the minimum travel cost, the maximum warehouse inventory and the maximum robustness.

[0022] Select two individuals from the mating pool as parents Parent1 and Parent2, and generate the crossover probability pc;

[0023] If p is less than pc, then: randomly select d∈[1,N] warehouses and select the route to which the dth warehouse of Parent1 and Parent2 belongs and Delete the parent1(Parent2) Duplicate client nodes, merge and As the new route R d And replace the routing in Parent1 (Parent2) To generate offspring Off1 and Off2; p is a random number between 0 and 1;

[0024] If p is greater than or equal to pc, then: randomly select two customer nodes v1 and v2 that do not belong to the same warehouse in each parent solution, and exchange the two customers to generate offspring;

[0025] Adjust the total warehouse inventory of the child. When the total warehouse inventory is less than the actual total demand, adjust the total warehouse inventory to the actual total demand; when the total warehouse inventory is greater than the actual total demand, adjust the total warehouse inventory according to the fixed percentage of the difference between the total warehouse inventory and the actual total demand.

[0026] Furthermore, the step of retraining the proxy model using the new offspring to update the proxy model specifically includes:

[0027] The true target value is used as the offspring solution set, individuals are selected from the offspring solution set and assigned to the proxy model, and the offspring solution set is added to the archive A used to store a fixed number of solutions. s middle;

[0028] When archive A s If the number of solutions is less than the maximum number of archives of the proxy model, the proxy model is retrained to update the proxy model;

[0029] Otherwise, from archive A s Select the number of solutions corresponding to the maximum number of archives, and according to the objective function values ​​corresponding to the selected solutions, select the solution corresponding to the optimal objective function value as the leader;

[0030] According to the true target value and prediction error, non-dominated sorting and crowding degree acquisition are performed, and the results of the first two levels of results are selected as elite solutions;

[0031] Get the Euclidean distance between each elite solution and the two characteristic parameters of the Leader; if the Euclidean distance is less than 40, get the next elite solution; otherwise, save it to the elite solution archive A e Add archive A s The elite solution whose Euclidean distance to the elite solution is less than 40; if A e If the number of solutions in is greater than 30, use A e Train the proxy model and s Remove the solution used for training and update the proxy model; if none of the above conditions are met, only update the proxy model.

[0032] Furthermore, the updating of the parent generation through the environment selection strategy specifically includes the following steps:

[0033] Merge the child and the parent to obtain a merged multiple path planning scheme; obtain the values ​​of the three objective functions of the upper model based on the merged multiple path planning scheme; obtain the solution using non-dominated sorting and congestion based on the values ​​of the three objective functions; select the path planning scheme corresponding to the solution that meets the requirements as the new parent.

[0034] Furthermore, the simulated annealing genetic algorithm is stopped when the termination condition is met, and the specific steps include:

[0035] The diversity of the population is obtained by using an entropy-based termination condition method;

[0036] The degree of change in population diversity is used as the termination condition. When the degree of diversity change is less than 0.001 in three consecutive generations of genetic algorithms, the algorithm iteration is stopped, otherwise it returns to the operation of crossover mutation on the parent generation to continue iteration.

[0037] An embodiment of the present invention provides a path planning system for multi-warehouse distribution tasks, including:

[0038] The upper model construction module is used to take the travel cost reflecting the vehicle's driving path cost, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model as the objective function; each customer is assigned to only one customer, each customer receives only one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive the delivery service of the same vehicle, the initial location and the end location of each vehicle's delivery route must be in the same warehouse, and sub-routes are eliminated as constraints, so as to construct a vehicle distribution task model that can generate multiple path planning solutions and use the vehicle distribution task model as the upper model of the two-layer optimization model;

[0039] A lower-layer model construction module is used to select the minimum travel cost from the travel costs of multiple path planning solutions randomly generated by the upper-layer model through simulated annealing, obtain the path planning solution corresponding to the minimum travel cost, and construct the lower-layer model of the two-layer optimization model;

[0040] The two-layer optimization model optimization module is used to use the multiple path planning schemes obtained according to the two-layer optimization model as the initial population of the simulated annealing genetic algorithm to train the Kriging model, obtain the path distance fitting model used to describe the customer allocation relationship and the path distance correlation, and use the path distance fitting model as the proxy model; perform crossover mutation on the multiple path planning schemes used as the initial population to generate offspring, and input the offspring into the proxy model to evaluate the offspring; according to the evaluation results, use non-dominated sorting to obtain the solution with the highest non-dominated level and obtain new offspring through local search method and simulated annealing method; use the new offspring to retrain the proxy model to update the proxy model and update the parent generation through the environment selection strategy; stop the simulated annealing genetic algorithm when the termination condition is met, and use the new offspring as the optimal planning path.

[0041] The embodiment of the present invention provides a path planning method and system for multi-warehouse distribution tasks. Compared with the prior art, the beneficial effects thereof are as follows:

[0042] A vehicle distribution task model that can generate multiple path planning schemes is constructed to reflect the path planning between multiple warehouses and multiple delivery customers. The method adopted is: the travel cost reflecting the vehicle driving path cost, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model are used as objective functions; each customer is assigned to only one customer, each customer only receives one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive the delivery service of the same vehicle, the initial location and end location of each vehicle's delivery route must be in the same warehouse, and the elimination of sub-routes are used as constraints. Then, multiple path planning schemes are obtained using the vehicle distribution task model and the minimum travel cost is selected from the multiple path planning schemes by simulated annealing. The path planning scheme corresponding to the minimum travel cost is obtained to complete the selection of the delivery path. The multiple path planning schemes obtained as the initial population by the two-layer optimization model are evaluated by the agent model, and the new offspring is obtained based on the evaluation results using the local search method and the simulated annealing method. The new offspring selects the optimal planning path from the multiple path planning schemes.

[0043] Among them, the path planning obtained by the vehicle distribution task model only considers the relationship between multiple warehouses and multiple distribution customers, without considering the impact of scenario factors other than warehouses and customers on path planning, so that path planning can adapt to a variety of scenarios, and through the agent model, local search method and simulated annealing method, the optimal path planning is selected from the path planning that can be applicable to a variety of scenarios, and finally the optimal planning path that can adapt to a variety of scenarios is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of an optimization method provided by an embodiment of the present invention;

[0045] Figure 2 An example diagram of a crossover operator provided by an embodiment of the present invention;

[0046] Figure 3 An example diagram of a mutation operator provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0048] See also Figure 1, an embodiment of the present invention provides a path planning method for a multi-warehouse distribution task, comprising the following steps:

[0049] Step 1: The path planning method is applied to a vehicle delivery task consisting of M warehouses, N task nodes with delivery time constraints, and K undifferentiated vehicles with capacity constraints.

[0050] Step 2: The travel cost reflecting the vehicle's driving path cost, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model are used as objective functions; each customer is assigned to only one customer, each customer receives only one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive the same delivery service, the initial and end locations of each vehicle's delivery route must be in the same warehouse, and sub-routes must be eliminated as constraints, so as to construct a vehicle distribution task model that can generate multiple path planning schemes and use the vehicle distribution task model as the upper model of the two-layer optimization model.

[0051] Step 3: Select the minimum travel cost from the travel costs of multiple path planning schemes randomly generated by the upper model through simulated annealing, obtain the path planning scheme corresponding to the minimum travel cost, and construct the lower model of the two-layer optimization model.

[0052] Step 4: Use the multiple path planning schemes obtained according to the two-layer optimization model as the initial population of the simulated annealing genetic algorithm to train the Kriging model, obtain the path distance fitting model used to describe the customer allocation relationship and the path distance correlation, and use the path distance fitting model as the proxy model; perform crossover mutation on the multiple path planning schemes as the initial population to generate offspring, and input the offspring into the proxy model to evaluate the offspring; based on the evaluation results, use non-dominated sorting to obtain the solution with the highest non-dominated level and obtain new offspring through local search method and simulated annealing method; use the new offspring to retrain the proxy model to update the proxy model and update the parent generation through the environmental selection strategy; stop the simulated annealing genetic algorithm when the termination condition is met, and use the new offspring as the optimal planning path.

[0053] 1. Establish a vehicle distribution task model.

[0054] 1.1. Define a node set consisting of M warehouse nodes and N distribution customer nodes:

[0055]

[0056] in, represents the mth warehouse node, represents the nth customer node, whose uncertainty demand is defined as dem i ; Define the path set E = {ei,j |i,j∈V},e i,j represents the edge connecting node i and node j, w i，j is the travel cost between node i and node j.

[0057] The scenario of demand change is defined as SC = {sc1, sc2, ..., sc t ,...,sc T}, sc t represents the tth scene, and T represents the total number of scenes.

[0058] 1.2. Use formulas (1), (2) and (3) to construct the objective function of the vehicle path planning model:

[0059]

[0060] The first objective F1 is the travel cost of the solution, which is also the goal of the lower-level optimization; the second objective F2 is the sum of all warehouse inventories, reflecting the inventory cost of the solution; the third objective F3 reflects the robustness of the solution.

[0061] in If the customer Assign to warehouse That otherwise, If there is a vehicle from the warehouse / customer Driving to warehouse / customer but otherwise in Represents warehouse In formula (3), SC is the scenario set, For the scene sc t Next customer demand, and Γ(true)=1, Γ(false)=0.

[0062] 1.3. Use equations (4) to (11) to construct the constraints of the vehicle path planning model:

[0063]

[0064] Where Q is the maximum load of the vehicle, u i Vehicle service customers Load capacity at the time.

[0065] Formula (4) ensures that each customer is assigned to only one customer; Formula (5) and Formula (6) are used to ensure that all customers are served only once; Formula (7) and Formula (8) prevent vehicles from traveling from one warehouse to another; Formula (9) ensures that customers assigned to different warehouses will not be served by the same vehicle; Formula (10) ensures that all vehicles start and end their routes at the same warehouse; Formula (11) aims to eliminate subroutes.

[0066] 2. Randomly initialize and generate an initial population of path planning solutions that meet the constraints, perform lower-level optimization, and train the proxy model.

[0067] 2.1. Randomly generate PopNum path planning solutions that meet the constraints as individuals in the initial population.

[0068] 2.2, in [dem i -ε*dem i ,dem i +ε*dem i ]A scene set SC is randomly generated within the interval.

[0069] 2.3. Calculate the lower layer solution f1 of the initial population according to formula (1) and perform lower layer optimization to obtain the lower layer optimal solution as F1. At the same time, obtain F2 and F3 according to formula (2) and formula (3).

[0070] 2.3.1. The lower layer optimization adopts simulated annealing, with the initial temperature of 1000, the stop temperature of 1, and the annealing rate (T) of the lower layer optimization in the initial population of 0.999.

[0071] 2.3.2. In each iteration, re-insertion, swap, and 2-Opt are randomly selected to generate offspring.

[0072] 2.3.3. Compare the optimal objectives of the offspring and the population. If the offspring is better, accept the offspring; otherwise, accept the bad solution with a certain probability.

[0073] 2.4. Add the initial population to archive A and train the proxy model.

[0074] 2.4.1. Extracting characteristic parameters of the solution in A and AGG={agg1,agg2,...,agg m},in represents the average distance from customers to the warehouse, The customer allocation plan is described from the perspective of customer aggregation, where A represents the area enclosed by customers and n represents the number of customers.

[0075] 2.4.2. Based on the features obtained in 2.4.1, the Kriging model is used for training to obtain the path distance fitting model, which is used as the proxy model.

[0076] 3. Generate offspring by crossing and mutating the population:

[0077] 3.1 Generate a mating pool of the gen-th generation with a size of .

[0078] 3.2 Select two individuals from the mating pool as parents Parent1 and Parent2, generate a random value pc, if p < pc, execute step 3.3; otherwise execute step 3.4; where pc is the crossover probability and p ∈ [0, 1].

[0079] 3.3 Randomly select a warehouse d ∈ [1, N], select the routes to which the d-th warehouses of Parent1 and Parent2 belong and delete the customer nodes in Parent1 (Parent2) that are repeated with , merge and as the new route R d and replace the routes in Parent1 (Parent2) to generate offspring Off1 and Off2.

[0080] As Figure 2 shown, the routes <b, 0, 6, 4, b> and <b, 3, 9, b> are selected from Parent1, and the routes <b, 0, 6, b> and <b, 8, 3, 9, b> are selected from Parent2; the node 8 that is repeated with the routes selected from Parent2 is deleted from Parent1, and similarly the node 4 that is repeated with the routes selected from Parent1 is deleted from Parent2; then, the remaining part of Parent1 and the selected part of Parent2 are merged into Offspring1, and similarly Offspring2 is synthesized; if there are missing customers in the offspring, they are assigned to the selected routes. For example, the customer 8 in Offspring2 is missing and is inserted into the route <b, 3, 9, b> to form the route <b, 3, 9, 8, b>.

[0081] 3.4 For each parent solution, randomly select two customer nodes v1 and v2 that do not belong to the same warehouse, and exchange the two customers to generate offspring Off.

[0082] As Figure 3 shown, customers 7 and 8 are exchanged, so the routes <a, 7, 1, a> and <b, 8, 3, 9, b> become <a, 8, 1, a> and <b, 7, 3, 9, b>.

[0083] 3.5 Check the inventory constraints of the generated offspring. If the total inventory is less than the total demand, adjust the inventory level to the total demand; otherwise, reduce the inventory level by 20% of the difference.

[0084] 4. Use the proxy model to evaluate offspring and select some offspring F for lower-level optimization:

[0085] 4.1. Extract the characteristic parameters of the offspring solution O and use the proxy model trained in the previous generation to evaluate the offspring.

[0086] 4.2. Perform non-dominated sorting based on the predicted value and prediction accuracy, locally search the first-level child solutions, and use a custom simulated annealing rate to perform lower-level optimization to obtain the true target value of the lower layer.

[0087] 4.2.1. Sort the predicted values ​​of the offspring solutions in ascending order, and set the initial T to Here K is the ranking level of the child solutions, that is, the better the target value, the more resources are allocated to the lower layer optimization.

[0088] 4.2.2. Check whether there is a sub-solution in A. If so, update T to Rep is the number of times the descendant solution has been optimized at the lower level, that is, the more times a solution appears, the more resources are allocated for optimization.

[0089] 4.2.3. Use formula (2) and formula (3) to calculate the upper target value of the offspring solution set O.

[0090] 5. Update the proxy model:

[0091] 5.1. Select the individual assigned to the proxy model s from the offspring solution set F for real evaluation and add it to A s middle.

[0092] 5.2 If A s If the number of solutions is less than the maximum number of archives ArchiveN, update s.

[0093] 5.3. Otherwise, select the individuals with ArchiveN before the target value as A s , and select the individual with the best target value as the leader.

[0094] 5.4. Perform non-dominated sorting and crowding calculation on F according to the true target value and prediction error, and select the individuals in the first two layers as the elite solution Elite.

[0095] 5.5. For each e∈Elite, calculate the Euclidean distance EucDis between it and the leader’s characteristic parameter. If the distance is less than 40, calculate the next individual; otherwise, go to A. e Add A s For individuals with EucDis less than 40 between A and e, e If the number of is greater than 30, use A e Training Agent Models e , while As Remove these individuals and re-update s; if none of the conditions are met, only update s without performing proxy model division.

[0096] 6. Update the next generation population through environmental selection strategy:

[0097] 6.1. Merge the offspring O and the parent P, perform non-dominated sorting and crowding calculation according to the three defined target values, and select the first PopNum individuals as the parent of the next iteration.

[0098] 7. If the termination condition is reached, the final population is output; otherwise, return to step 3 to execute:

[0099] 7.1. Adopt the termination condition based on entropy, and the entropy is calculated as follows:

[0100]

[0101] Among them, H represents the diversity of the population.

[0102] 7.2. If the diversity change degree of a population is H n If it is less than 0.001 for three consecutive generations, it is considered that the algorithm is difficult to find a better solution, and the algorithm iteration is stopped; otherwise, it jumps to step 3 to continue. n The calculation is as follows:

[0103]

[0104] in, is the theoretical minimum value of H, is the theoretical maximum value.

[0105] The beneficial effects of the present invention include the following:

[0106] 1. This paper proposes a proxy model tailored for the multi-warehouse vehicle distribution problem to predict the minimum routing cost of customer allocation plans. Specifically, in order to solve the discrete decision variables of the multi-warehouse vehicle distribution problem, two sets of continuous features are extracted from the customer allocation plans as model inputs. The first set of features reflects the degree of customer aggregation in each warehouse, while the second set of features describes the distance between customers and warehouses. These two sets of features can comprehensively characterize the customer allocation plans and help the proxy model effectively distinguish the advantages and disadvantages of the plans.

[0107] 2. The present invention proposes a special model management strategy, which adaptively divides the decision space of the customer allocation problem into different subspaces during the evolutionary algorithm search process, and then constructs proxy models in different subspaces. Since each proxy model only needs to approximate the relationship between the characteristics of the customer allocation solution and the minimum travel cost in a very small decision space, it can effectively alleviate the performance bottleneck caused by multiple local optima. The number of local optima in a given subspace is significantly less than the number of optima in the entire decision space.

[0108] 3. The present invention proposes a simulated annealing algorithm to search for the optimal route of each vehicle segment. In order to solve the problem that the inaccurate optimal path leads to the degradation of the proxy model performance, an adaptive computing resource allocation strategy is introduced into the simulated annealing algorithm to allocate more computing resources to potential customer allocation plans; thereby providing a more accurate evaluation of these allocation plans, so that the algorithm can find a better route and continuously improve the accuracy of the model.

[0109] An embodiment of the present invention provides a path planning system for multi-warehouse distribution tasks, including:

[0110] The upper model construction module is used to take the travel cost reflecting the vehicle's driving path cost, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model as objective functions; each customer is assigned to only one customer, each customer receives only one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive the same delivery service, the initial location and end location of each vehicle's delivery route must be in the same warehouse, and sub-routes are eliminated as constraints, so as to construct a vehicle distribution task model that can generate multiple path planning solutions and use the vehicle distribution task model as the upper model of the two-layer optimization model.

[0111] The lower model construction module is used to select the minimum travel cost from the travel costs of multiple path planning schemes randomly generated by the upper model through simulated annealing, obtain the path planning scheme corresponding to the minimum travel cost, and construct the lower model of the two-layer optimization model.

[0112] The two-layer optimization model optimization module is used to use the multiple path planning schemes obtained according to the two-layer optimization model as the initial population of the simulated annealing genetic algorithm to train the Kriging model, obtain the path distance fitting model used to describe the customer allocation relationship and the path distance correlation, and use the path distance fitting model as the proxy model; perform crossover mutation on the multiple path planning schemes used as the initial population to generate offspring, and input the offspring into the proxy model to evaluate the offspring; according to the evaluation results, use non-dominated sorting to obtain the solution with the highest non-dominated level and obtain new offspring through local search method and simulated annealing method; use the new offspring to retrain the proxy model to update the proxy model and update the parent generation through the environment selection strategy; stop the simulated annealing genetic algorithm when the termination condition is met, and use the new offspring as the optimal planning path.

[0113] A specific embodiment is as follows:

[0114] The present invention aims to solve the path planning problem in uncertain transportation environments, such as demand changes caused by garbage collection, uncertain demand and starting from multiple distribution stations, and use the proxy model to assist the two-layer optimization algorithm to perform lower-level path planning, reduce its consumption of computing resources, and at the same time, through a special model management strategy, continuously improve the accuracy of the model, balance the robustness and optimality of the population, so as to obtain the best robust optimal solution in various scenarios. Specifically, the path planning method is applied to a vehicle distribution task consisting of M warehouses, N task nodes with delivery time constraints and K undifferentiated vehicles with capacity constraints, and is carried out in the following steps:

[0115] Step 1: Establish a vehicle delivery task model.

[0116] Step 2: Randomly initialize and generate an initial population of path planning solutions that meet the constraints, perform lower-level optimization, and train the proxy model.

[0117] Step 3: Perform crossover mutation on the population to generate offspring.

[0118] Step 4: Use the proxy model to evaluate the offspring and select some offspring F for lower-level optimization.

[0119] Step 5: Update the proxy model.

[0120] Step 6: Update the next generation population through environmental selection strategy.

[0121] Step 7: If the termination condition is met, output the final population; otherwise, return to step 3.

[0122] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A path planning method for multi-warehouse distribution tasks, characterized in that: The path planning method is applied to a vehicle delivery task consisting of M warehouses, N task nodes with delivery time constraints, and K indifferent vehicles with capacity constraints; the path planning method comprises the following steps: The travel cost reflecting the cost of the vehicle's driving path, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model are used as objective functions; each customer is assigned to only one customer, each customer receives only one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive the delivery service of the same vehicle, the initial location and the end location of each vehicle's delivery route must be in the same warehouse, and sub-routes must be eliminated as constraints, so as to construct a vehicle distribution task model that can generate multiple path planning solutions and use the vehicle distribution task model as the upper model of the two-layer optimization model; The minimum travel cost is selected from the travel costs of multiple path planning schemes randomly generated by the upper model through simulated annealing, and the path planning scheme corresponding to the minimum travel cost is obtained to construct the lower model of the two-layer optimization model; The multiple path planning schemes obtained from the two-layer optimization model are used as the initial population of the simulated annealing genetic algorithm to train the Kriging model, and the path distance fitting model used to describe the correlation between customer allocation relationship and path distance is obtained, and the path distance fitting model is used as the proxy model. A plurality of path planning schemes as the initial population are subjected to crossover mutation to generate offspring, and the offspring are input into the proxy model for evaluation; based on the evaluation results, the non-dominated sorting is used to obtain the solution with the highest non-dominated level, and new offspring are obtained through the local search method and the simulated annealing method; the proxy model is retrained with the new offspring to update the proxy model and the parent generation is updated through the environment selection strategy; the simulated annealing genetic algorithm is stopped when the termination condition is met, and the new offspring is used as the optimal planning path.

2. A path planning method for multi-warehouse distribution tasks as claimed in claim 1, characterized in that: The method of selecting the minimum travel cost from the travel costs of multiple path planning solutions randomly generated by the upper model by simulated annealing includes the following specific steps: Randomly generate multiple path planning solutions that meet the objective function and constraints; Obtaining solution results of the first objective function, the second objective function and the third objective function of the initial population; Designing a simulated annealing method, including: setting the initial temperature to 1000 degrees Celsius, the stop temperature to 1 degree Celsius, and the annealing rate to 0.999 degrees Celsius / second; Using a simulated annealing method to process the solution results of the first objective function, the second objective function and the third objective function to obtain a final solution; In the iterative process of the simulated annealing, a local search heuristic algorithm optimized by reinsertion, Swap or 2-Opt is used to generate offspring, and a value suitable for the objective function is selected from the offspring and the final solution to update the final solution.

3. A path planning method for multi-warehouse distribution tasks as claimed in claim 1, characterized in that: The step of obtaining a path distance fitting model for describing the correlation between the customer allocation relationship and the path distance, and using the path distance fitting model as a proxy model, specifically comprises the following steps: According to the multiple path planning schemes as the initial population, two characteristic parameters of each scheme are extracted; wherein the two characteristic parameters are the average distance from the customer to the warehouse and the aggregation degree of the customer allocation scheme; The Kriging model is trained based on the two characteristic parameters to obtain the proxy model.

4. A path planning method for multi-warehouse distribution tasks as claimed in claim 1, characterized in that: The specific steps of performing crossover mutation on the multiple path planning schemes as the initial population to generate offspring include: Based on multiple path planning schemes as the initial population, a binary bidding method is used to generate a The mating pool contains the path planning scheme corresponding to the minimum travel cost, the maximum warehouse inventory and the maximum robustness. Select two individuals from the mating pool as parents Parent1 and Parent2, and generate the crossover probability pc; If p is less than pc, then: randomly select d∈[1,N] warehouses and select the route to which the dth warehouse of Parent1 and Parent2 belongs and Delete the Parent1(Parent2) Duplicate client nodes, merge and As the new route R d And replace the routing in Parent1 (Parent2) To generate offspring Off1 and Off2; p is a random number between 0 and 1; If p is greater than or equal to pc, then: randomly select two customer nodes v1 and v2 that do not belong to the same warehouse in each parent solution, and exchange the two customers to generate offspring; Adjust the total warehouse inventory of the child. When the total warehouse inventory is less than the actual total demand, adjust the total warehouse inventory to the actual total demand; when the total warehouse inventory is greater than the actual total demand, adjust the total warehouse inventory according to the fixed percentage of the difference between the total warehouse inventory and the actual total demand.

5. A path planning method for multi-warehouse distribution tasks as claimed in claim 1, characterized in that: The step of retraining the proxy model using the new offspring to update the proxy model includes: The true target value is used as the offspring solution set, individuals are selected from the offspring solution set and assigned to the proxy model, and the offspring solution set is added to the archive A used to store a fixed number of solutions. s middle; When archive A s If the number of solutions is less than the maximum number of archives of the proxy model, the proxy model is retrained to update the proxy model; Otherwise, from archive A s Select the number of solutions corresponding to the maximum number of archives, and according to the objective function values ​​corresponding to the selected solutions, select the solution corresponding to the optimal objective function value as the leader; According to the true target value and prediction error, non-dominated sorting and crowding degree acquisition are performed, and the results of the first two levels of results are selected as elite solutions; Get the Euclidean distance between each elite solution and the two characteristic parameters of the Leader; if the Euclidean distance is less than 40, get the next elite solution; otherwise, save it to the elite solution archive A e Add archive A s The elite solution whose Euclidean distance to the elite solution is less than 40; if A e If the number of solutions in is greater than 30, use A e Train the proxy model and s Remove the solution used for training and update the proxy model; if none of the above conditions are met, only update the proxy model.

6. A path planning method for multi-warehouse distribution tasks as claimed in claim 1, characterized in that: The specific steps of updating the parent generation through the environment selection strategy include: Merge the child and the parent to obtain multiple path planning solutions; Obtain the values ​​of the three objective functions of the upper model according to the combined multi-path planning scheme; According to the values ​​of the three objective functions, the solution is obtained using non-dominated sorting and crowding; Select the path planning solution corresponding to the solution that meets the requirements as the new parent.

7. A path planning method for multi-warehouse distribution tasks as claimed in claim 1, characterized in that: The simulated annealing genetic algorithm is stopped when the termination condition is met, and the specific steps include: The diversity of the population is obtained by using an entropy-based termination condition method; According to the degree of change in the diversity of the population, when the degree of change in diversity is less than 0.001 in three consecutive generations of genetic algorithms, the algorithm iteration is stopped, otherwise it returns to the operation of crossover mutation on the parent generation to continue iteration.

8. A path planning system for multi-warehouse distribution tasks, characterized in that: include: An upper model building module is used to take the travel cost reflecting the vehicle's driving path cost, the total warehouse inventory reflecting the inventory cost, and the robustness of the vehicle distribution task model as the objective function; Each customer is assigned to only one customer, each customer receives only one delivery service, each vehicle cannot travel from one warehouse to another, customers assigned to different warehouses cannot receive delivery services from the same vehicle, the starting and ending locations of each vehicle's delivery route must be in the same warehouse, and sub-routes are eliminated as constraints to build a vehicle delivery task model that can generate multiple path planning solutions and use the vehicle delivery task model as the upper model of the two-layer optimization model; A lower-layer model construction module is used to select the minimum travel cost from the travel costs of multiple path planning solutions randomly generated by the upper-layer model through simulated annealing, obtain the path planning solution corresponding to the minimum travel cost, and construct the lower-layer model of the two-layer optimization model; The two-layer optimization model optimization module is used to use the multiple path planning schemes obtained according to the two-layer optimization model as the initial population of the simulated annealing genetic algorithm to train the Kriging model, obtain the path distance fitting model used to describe the customer allocation relationship and the path distance correlation, and use the path distance fitting model as the proxy model; perform crossover mutation on the multiple path planning schemes used as the initial population to generate offspring, and input the offspring into the proxy model to evaluate the offspring; according to the evaluation results, use non-dominated sorting to obtain the solution with the highest non-dominated level and obtain new offspring through local search method and simulated annealing method; use the new offspring to retrain the proxy model to update the proxy model and update the parent generation through the environment selection strategy; stop the simulated annealing genetic algorithm when the termination condition is met, and use the new offspring as the optimal planning path.

Images