A route planning method and system for large-scale vehicle routing problem

Abstract

The application discloses a route planning method and system for a large-scale vehicle path problem, belongs to the technical field of vehicle path planning, groups multiple task nodes, distributes the task nodes to corresponding groups, forms multiple populations of an upper-layer clustering scheme from the multiple groups, determines the number of vehicles and the total length of paths corresponding to each individual in the population corresponding to each group, performs non-dominated sorting on each individual, determines the non-dominated sorting level of each individual, uses the non-dominated sorting level as a label to construct a training sample of an SVM model, inputs the generated offspring population into the trained SVM model, obtains the classification result of each individual in the offspring population, selects excellent offspring, performs lower-layer path optimization, obtains the optimal route corresponding to the excellent offspring, and selects a new generation of population; the iteration is performed until the new generation of population meets a preset termination condition, and an optimal route planning scheme is output. The method can reduce the time complexity in the route planning process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle path planning, and more particularly to a route planning method and system for large scale vehicle routing problem. BACKGROUND

[0002] Vehicle Routing Problem (VRP) is an important problem in the field of operations research and logistics management, which involves how to effectively plan the delivery routes of multiple vehicles to meet customer demand while minimizing transportation cost and time. VRP is a typical NP-hard problem, which means that as the problem size increases, the solution time will increase dramatically. Large Scale Vehicle Routing Problem (LSVRP) is an extension of vehicle routing problem, which involves a large number of customer points and vehicles, making the problem more complex and difficult to solve. As the problem size increases, traditional heuristic algorithms and exact algorithms face great challenges in computation time and resource consumption.

[0003] Currently, the traditional route planning method for large scale vehicle routing problem mainly includes heuristic algorithms, exact algorithms and machine learning methods. Among them, heuristic algorithms such as ant colony algorithm, genetic algorithm, simulated annealing, tabu search, etc. These algorithms can give approximate optimal solution within acceptable time, but may fall into local optimal solution and the convergence speed may be slow. Exact algorithms such as branch and bound method and dynamic programming method, etc. These methods can get optimal solution on small scale problem, but become impractical on large scale problem due to the increase of computational complexity. Machine learning methods, by learning the patterns in data, can provide new solutions for VRP problem, but may have insufficient generalization ability on different data distributions when dealing with large scale data.

[0004] In summary, the route planning method for large scale vehicle routing problem in the prior art, due to the large scale vehicle routing problem involving a large number of customer points and vehicles, will lead to an increase in computational complexity when dealing with large scale data, and there is insufficient generalization ability on different data distributions, which affects the obtaining of optimal solution. SUMMARY

[0005] In view of the problems in the above field, the application provides a route planning method and system for large-scale vehicle routing problems, which allocates task nodes to corresponding groups through clustering to find a more suitable clustering scheme, and can effectively overcome the problem of unsatisfactory clustering results caused by improper selection of initial centroid and clustering number. The paths of the selected new generation population are optimized to obtain the corresponding optimal path, and a double-layer optimization framework is used to find a path planning scheme with shorter distance and lower cost. By selecting a new generation population and filtering out individuals with better performance, the upper clustering scheme is filtered, which can well solve the large-scale vehicle routing problem with capacity constraints, effectively reduce the problem complexity, and further strengthen the reliability of the path planning.

[0006] To solve the above technical problems, the application discloses a route planning method for large-scale vehicle routing problems, comprising the following steps:

[0007] Obtaining a plurality of task nodes of a large-scale vehicle routing problem;

[0008] Grouping the plurality of task nodes, determining the clustering center of each group and the number of task node groups, and when the vehicle distribution task corresponding to the group to which the clustering center belongs meets the vehicle capacity limit requirement, allocating the task node to the corresponding group, and forming the corresponding population of the upper clustering scheme for each group;

[0009] Determining the number of vehicles and the total length of the path corresponding to each individual in the population corresponding to each group, non-dominantly sorting each individual, determining the non-dominant sorting level of each individual as a label to construct a training sample, and inputting the training sample into an SVM model to train the model;

[0010] Generating a child population by performing a differential mutation operation on the current population;

[0011] Inputting the child population into the trained SVM model to obtain the classification result of each individual in the child population, selecting excellent children and performing lower path optimization to obtain the optimal route corresponding to the excellent children, merging the current population and the excellent children, non-dominantly sorting in the upper layer to select a new generation population, and iterating until the new generation population meets the preset termination condition, and outputting the optimal route planning scheme.

[0012] Preferably, the determination of the clustering center of each group and the number of task node groups comprises the following steps:

[0013] Determining the current generation number gen=1, and the gen generation population:

[0014]

[0015] wherein, a cluster scheme of the num-th individual of the gen-th generation, the number of cluster centers of the num-th individual of the gen-th generation and whether the i-th task node is selected as a cluster center, indicates that the i-th task node is selected as a cluster center, indicates that the i-th task node is not selected as a cluster center, N u indicates the population size;

[0016] initializing population grouping information indicates the grouping information of the num-th individual of the gen-th generation;

[0017] according to initializing individual task node grouping Group, obtaining:

[0018]

[0019] wherein, g represents the number of task node groups, and

[0020] Preferably, the task nodes are distributed to corresponding groups, and each group forms a corresponding population of the upper cluster scheme, and specifically comprises:

[0021] determining the minimum grouping number C according to the ratio of the task demand and the vehicle load, and sorting the tasks according to the demand from large to small;

[0022] clustering the task nodes according to the cluster scheme individuals generated by the upper layer, and preferentially distributing the task nodes with larger demand during task node distribution;

[0023] Step 3.1: initializing num = 1;

[0024] Step 3.2: obtaining the task grouping scheme of the current individual

[0025] Step 3.3: initializing index = 1;

[0026] Step 3.4: calculating the distance between the task node index and each cluster center, and simultaneously judging whether the remaining capacity of the group j to which the task node index is distributed meets the capacity limit requirement of the warehouse after the task node index is distributed to the group j with the smallest distance, and when the capacity limit requirement of the vehicle is met, the task node index is distributed to the group j, and tc is recorded in TaskGroup; otherwise, the task node index is distributed to the group j with the suboptimal distance, and whether the group j meets the capacity limit requirement of the vehicle is judged. index =j; otherwise, the task node index is distributed to the group j with the suboptimal distance, and whether the group j meets the capacity limit requirement of the vehicle is judged. next meets the capacity limit requirement of the vehicle.next Given the remaining capacity, repeat this step until the task node index is allocated and j = j next ;

[0027] Step 3.5: Set index = index + 1. When index is less than or equal to L, execute the allocation process center for the index-th task node, where L represents the number of delivery task nodes. When index is greater than L, it proves that all task nodes of the num-th individual have been assigned to groups. Update the task grouping scheme for the current individual using the group information of each task node recorded in TaskGroup.

[0028] Step 3.6: Take num = num + 1, when num is less than or equal to N u When num is greater than N, proceed to step 3.2; u When the time is reached, it indicates that all individuals in the current population have completed the grouping process of the task node, and the iteration terminates.

[0029] Preferably, determining the number of vehicles and the total path length corresponding to each individual in the population of each group includes the following steps:

[0030] Step 4.1: Initialize num = 1;

[0031] Step 4.2: Obtain the task grouping scheme for the current individual.

[0032] Step 4.3: Initialize j = 1;

[0033] Step 4.4: Execute the lower-level optimization process for the current group to obtain the current group. k represents the number of task nodes in the current group. The coordinates of the warehouse and the task nodes in this group are obtained. The optimal travel path is found using the ant colony algorithm. The 2-opt algorithm is then used to perform local optimization on the optimal travel path to obtain the task node access order (TaskOrder). num ={Order1, Order2,..., Order k} and total path length Dist num ;

[0034] Step 4.5: Update the total path length of the current task group and node access order

[0035] Step 4.6: Let j = j + 1. When j <= g, execute the lower-level optimization process for the j-th group; when j > g, it means that the task grouping scheme of this individual has completed the lower-level optimization process, and update the total path length of this individual. Number of individual vehicles and node access order

[0036] Step 4.7: Let num = num + 1. When num is less than or equal to N... u When num is greater than N, proceed to step 4.2; u When the iteration ends, it indicates that all individuals in the current population have completed the lower-level optimization process.

[0037] Preferably, training the model specifically includes:

[0038] Get the total path length of each individual in the current population. and the number of vehicles required

[0039] Based on the total path length and number of vehicles for each individual, determine the non-dominated ranking level for each individual. in, This represents the non-dominant ranking rank of the i-th individual in the gn-th generation; Greater than or equal to 1 Less than or equal to N u ;

[0040] Set the label of individuals at level 1 to 1, and set the labels of individuals at other levels to -1 to obtain the labels of the current population.

[0041] Obtain the clustering scheme of the population This represents the num-th individual in the gn-th generation;

[0042] Clustering scheme P gen and Tags gen The input is fed into the SVM model to train the model.

[0043] Preferably, the generation of offspring population specifically includes:

[0044] Step 6.1: Initialize the offspring population

[0045] Step 6.2: Initialize num = 1;

[0046] Step 6.3: Obtain the clustering scheme for the current individual. Randomly select a non-dominant ranking level Clustering scheme for individuals Randomly selected from this clustering scheme Obtain an individual and res is not equal to num;

[0047] Step 6.4: Perform the difference operation to obtain individuals in the offspring population after the difference operation:

[0048]

[0049] wherein F is a difference scaling factor;

[0050] Step 6.5: Determine whether the variables in the individuals in the offspring population after the difference operation exceed the limits, and modify the variable value to the upper limit when greater than the upper limit, or to the lower limit when less than the lower limit;

[0051] Step 6.6: Perform a polynomial mutation operation on the variables in the individuals in the offspring population after the difference operation, and again check whether the variables after the operation meet the requirements, and modify the variables after the mutation operation according to step 6.5 if they do not meet the requirements;

[0052] Step 6.7: Let num=num+1, and determine whether num>N u is true, which indicates that the gennth offspring population is obtained Otherwise, step 6.3 is performed.

[0053] Preferably, the optimal route corresponding to the excellent offspring is obtained, specifically comprising:

[0054] The offspring population Childgen is input into the trained SVM model to obtain the offspring population label ChildTag

[0055] The number of labels greater than 0 in ChildTag gen is calculated, and the corresponding index ChildIndex gen ={i1, i2,..., i k} is obtained, and when the number is greater than or equal to SubCount, the next operation is performed; when the number is less than SubCount, the individuals that perform well and whose predicted results are negative classes are selected to be filled in, ChildIndex gen ={i1, i2,..., i k , j1,..., j l}, k+l==SubCount;

[0056] The corresponding excellent offspring is obtained through the index ChildIndex gen of the excellent offspring:

[0057]

[0058] The excellent offspring Child​​gen performing a lower layer optimization process when all individuals in the excellent child gen obtaining the number of vehicles and the total length of paths corresponding to the excellent child when the lower layer optimization process is completed.

[0059] Preferably, the selecting a new generation population comprises the following steps:

[0060] merging the parent population P gen and the child population Child gen into a temporary population Pop gen ;

[0061] The size of the temporary population is denoted as PopCount, and the temporary population is non-dominantly sorted according to the two target values of the total length of paths and the number of vehicles, and the non-dominant sorting level and the crowding distance of each individual are calculated.

[0062] The crowding distance calculation formula is as follows:

[0063]

[0064] where i is the ith individual in the temporary population, i < PopCount, M is the target number, j = 0 indicates that the target is the number of vehicles, j = 1 indicates that the target is the total length of paths traveled by the vehicles, when the non-dominant level is the same, the crowding distance of the ith solution is the sum of the differences between the two adjacent solutions in each dimension target, and the crowding distances of the first and last individuals are set to infinity.

[0065] The temporary population is screened according to the non-dominant level and the crowding distance, and the individuals on the first front face are screened, when the screened first front face individuals exceed N u , N u individuals are screened by the crowding distance, otherwise all individuals on the front face are selected; when the screened first front face individuals are less than N u , the individuals on the remaining front face are selected according to the rule until the selection is completed, and the final population P gen+1 is output as the new generation population.

[0066] The output new generation population P gen+1 is input into the SVM model, and the model is updated.

[0067] Preferably, the output optimal route planning scheme comprises the following steps:

[0068] According to the new generation population P gen+1, take gen = gen + 1, judge whether the current iteration algebra gen is greater than the preset population algebra Gen, when the preset population algebra Gen condition is not satisfied, continue to execute the iterative optimization process; when the condition is satisfied, calculate the hyper volume HV value of the nearest Gen generation population, that is Obtain the maximum and minimum values in HVgen and record them as hv max and hv min ;

[0069] By calculating the diversity change degree H l of the new generation population:

[0070]

[0071] When H l is greater than the preset population diversity change threshold, continue to execute the iterative optimization process; otherwise, randomly select an individual with a dominance level IndividualLevel = 1 from the current new generation population, and its index is index, obtain its optimal route scheme TaskOrder index , and output the number of vehicles and the total length of the path corresponding to the optimal path of the new generation population as the optimal route planning scheme.

[0072] Preferably, the route planning system for large-scale vehicle routing problem comprises:

[0073] A task node acquisition module is configured to acquire a plurality of task nodes of the large-scale vehicle routing problem;

[0074] An upper layer clustering module is configured to group the plurality of task nodes, determine a clustering center of each group and a number of task node groups, and when the vehicle distribution task corresponding to the group to which the clustering center belongs satisfies the vehicle capacity limit requirement, allocate the task node to the corresponding group, and form a corresponding population of the upper layer clustering scheme for each group;

[0075] A lower layer path optimization module is configured to determine the number of vehicles and the total length of the path corresponding to each individual in the corresponding population of each group, non-dominantly sort each individual, determine the non-dominantly sorted level of each individual as a label to construct a training sample, input the training sample into an SVM model, train the model, generate a child population by performing a difference mutation operation on the current population, input the child population into the trained SVM model, obtain the classification result of each individual in the child population, select excellent children and perform lower layer path optimization to obtain the optimal route corresponding to the excellent children, and merge the current population and the excellent children, non-dominantly sort in the upper layer to select a new generation population. Iteratively execute until the new generation population satisfies the preset termination condition, and output the optimal route planning scheme.

[0076] Compared with the prior art, the present application has the following beneficial effects:

[0077] The route planning method for the large-scale vehicle routing problem provided by the present application can effectively overcome the problem of unsatisfactory clustering results caused by improper selection of initial centroids and clustering numbers by converting the actual path optimization problem into a bi-level optimization problem, assigning task nodes to corresponding groups through clustering in the upper layer of the bi-level optimization framework, and finding a more suitable clustering scheme. The number of vehicles and the total length of the path corresponding to each individual in the lower layer path optimization of each individual in the population corresponding to each group are determined, a training sample is constructed to train the SVM model, and the SVM model after training is used to select potential individuals from the offspring population to perform lower layer optimization. The generated offspring population is input into the trained SVM model to divide each individual in the offspring population into positive and negative classes, select excellent offspring and perform lower layer path optimization to obtain the optimal route corresponding to the excellent offspring, and merge the current population and the excellent offspring, perform non-dominated sorting in the upper layer, select a new generation population, and update the SVM model. When the selected new generation population meets the preset termination condition, the optimal route planning scheme is output. The method optimizes the path of the new generation population to obtain the corresponding optimal path, and uses the bi-level optimization framework to find a path planning scheme with shorter distance and lower cost. By selecting the new generation population, filtering out individuals with better performance, and filtering the upper layer clustering scheme, the large-scale vehicle routing problem with capacity constraints can be well solved, the problem complexity is effectively reduced, and the reliability of the route planning can be further strengthened. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 A flowchart of the route planning method for the large-scale vehicle routing problem provided by the present application is shown in the figure.

[0079] Figure 2 A bi-level optimization framework in the route planning method provided by the present application is shown in the figure.

[0080] Figure 3 A flowchart of dividing each individual in the offspring population into positive and negative classes provided by the embodiment of the present application is shown in the figure.

[0081] Figure 4 An example diagram of the selection strategy of the upper layer clustering and lower layer optimization provided by the embodiment of the present application is shown in the figure.

[0082] Figure 5 An example diagram of the strategy of generating a new generation population provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0083] The above description is only a specific implementation of the present application. The present application is not limited to the above embodiments. Any changes, modifications, equivalents, improvements, and the like within the spirit and principle of the present application shall fall within the scope of the present application. Figures 1-5The technical solutions in the embodiments of the present application are clearly and completely described. It should be understood that the terms described in the present application are only for describing the specific embodiments, and are not used to limit the present application.

[0084] As shown in Figure 1 The present application proposes a route planning method for large-scale vehicle routing problem, comprising the following steps:

[0085] Step 1: obtaining a plurality of task nodes of a large-scale vehicle routing problem;

[0086] Step 2: grouping the plurality of task nodes, determining the cluster center of each group and the number of task node groups; when the cluster center belongs to the group corresponding to the vehicle distribution task satisfying the vehicle capacity limit requirement, the task node is allocated to the corresponding group, and each group forms the corresponding population of the upper layer clustering scheme;

[0087] Step 3: performing lower path optimization on each group to determine the number of vehicles and the total length of the path corresponding to each individual; according to the number of vehicles and the total length of the path corresponding to each individual, the non-dominated sorting of each individual is performed and the sorting result is labeled to construct a training sample; input the training sample into the SVM model, and train the model;

[0088] Step 4: determining the number of vehicles and the total length of the path corresponding to each individual in the population corresponding to each group, performing non-dominated sorting on each individual, determining the non-dominated sorting level of each individual as a label to construct a training sample; input the training sample into the SVM model, and train the model;

[0089] Step 5: generating a child population by performing differential mutation operation on the current population; inputting the child population into the trained SVM model to obtain the classification result of each individual in the child population, selecting excellent children and performing lower path optimization to obtain the optimal route corresponding to the excellent children; by merging the current population and the excellent children, non-dominated sorting is performed in the upper layer to select a new population; iterate until the new population meets the preset termination condition, output the optimal route planning scheme, as shown in Figure 2 .

[0090] Specifically, in step S2, the cluster center of each group and the number of task node groups are determined, which specifically includes the following steps:

[0091] Determine the current generation number gen = 1, the gen generation population:

[0092]

[0093] Wherein, The cluster scheme of the num-th individual in the gen-th generation is represented by the number of cluster centers representing the num-th individual of the gen-th generation and whether the i-th task node is selected as a cluster center, representing that the i-th task node is selected as a cluster center, representing that the i-th task node is not selected as a cluster center, N u representing the population size.

[0094] initializing population grouping information representing the grouping information of the num-th individual of the gen-th generation.

[0095] according to initializing individual task node grouping Group, obtaining:

[0096]

[0097] wherein, g represents the number of task node groups, and

[0098] allocating the task nodes to corresponding groups and forming each group into a corresponding population of the upper layer clustering scheme, specifically including:

[0099] determining the minimum group number C according to the ratio of the task demand quantity and the vehicle load, and sorting the tasks according to the demand quantity from large to small;

[0100] grouping the task nodes according to the upper layer generated clustering scheme individual, and preferentially allocating the task nodes with larger demand quantity when allocating the task points;

[0101] Step 2.1: initializing num = 1;

[0102] Step 2.2: obtaining the task grouping scheme of the current individual

[0103] Step 2.3: initializing index = 1;

[0104] Step 2.4: calculating the distance between the task node index and each cluster center, and simultaneously judging whether to allocate the task node index to the group with the smallest distance, then judging whether the remaining capacity of the group meets the capacity limit requirement of the warehouse, when the capacity limit requirement of the vehicle is met, allocating the task node index to the group j and recording tc in TaskGroup index =j; otherwise, allocating the task node index to the suboptimal cluster j next and judging whether the remaining capacity of the group j next meets the requirement, repeating the step until the allocation of the task node index is completed and j=j next ;

[0105] Step 2.5: Take index=index+1, when index is less than or equal to L, execute the allocation process center of the index task node, wherein L represents the number of distribution task nodes; when index is greater than L, it is proved that all task nodes of the num individual have been allocated into groups, and the task grouping scheme of the current individual is updated through the group information recorded in the TaskGroup of each task node belonging to the group

[0106] Step 2.6: Take num=num+1, when num is less than or equal to N u , execute step 2.2; when num is greater than N u , it is proved that all individuals in the current population have completed the grouping process of task nodes, and the iteration is terminated.

[0107] In step S3, the number of vehicles and the total length of the path corresponding to each individual are determined, including the following steps:

[0108] Step S3.1: Initialize num=1;

[0109] Step S3.2: Obtain the task grouping scheme of the current individual

[0110] Step S3.3: Initialize j=1;

[0111] Step S3.4: Execute the lower-level optimization process of the current group, and obtain the task nodes in the current group k represents the number of task nodes in the current group, obtain the coordinates of the warehouse and the task nodes in this group, and seek the optimal driving path through the ant colony algorithm; through the 2-opt algorithm, the optimal driving path is locally optimized to obtain the access order TaskOrder num ={Order1, Order2,..., Order k} and the total length of the path Dist num ;

[0112] Step S3.5: Update the total length of the path and the node access order

[0113] Step S3.6: Let j=j+1, when j<=g, execute the lower-level optimization process of the jth group; when j>g, it is proved that the task grouping scheme of the individual has completed the lower-level optimization process, and the total length of the path the number of vehicles of the individual and the node access order

[0114] Step S3.7: Let num = num + 1, when num is less than or equal to N u When num is greater than N, proceed to step S3.2; u When the iteration ends, it indicates that all individuals in the current population have completed the lower-level optimization process.

[0115] Training the model specifically includes:

[0116] Get the total path length of each individual in the current population. and the number of vehicles required

[0117] Based on the total path length and number of vehicles for each individual, determine the non-dominated ranking level for each individual. in, This represents the non-dominant ranking rank of the i-th individual in the gen-th generation; Greater than or equal to 1 Less than or equal to N u ;

[0118] Set the label of individuals at level 1 to 1, and set the labels of individuals at other levels to -1 to obtain the labels of the current population.

[0119] Obtain the clustering scheme of the population This represents the num-th individual in the gn-th generation;

[0120] Clustering scheme p gen and Tags gen The input is fed into the SVM model to train the model.

[0121] In step S4, the offspring population is generated, specifically including:

[0122] Step S4.1: Initialize the offspring population

[0123] Step S4.2: Initialize num = 1;

[0124] Step S4.3: Obtain the clustering scheme for the current individual. Randomly select a non-dominant ranking level Clustering scheme for individuals Randomly selected from this clustering scheme Obtain an individual And res is not equal to num;

[0125] Step S4.4: performing a difference operation to obtain individuals in the offspring population after the difference operation:

[0126]

[0127] wherein F is a difference scaling factor;

[0128] Step S4.5: judging whether the variables in the individuals in the offspring population after the difference operation exceed the limits, modifying the variable value to the upper limit when greater than the upper limit, and modifying the variable value to the lower limit when less than the lower limit;

[0129] Step S4.6: performing a polynomial mutation operation on the and again checking whether the variables after the operation meet the requirements, and modifying the variables after the mutation operation according to step S4.5 when they do not meet the requirements;

[0130] Step S4.7: setting rum=num+1, judging whether num>N u is true, indicating that the gernth offspring population is obtained, otherwise, step S4.3 is performed.

[0131] Obtaining the optimal route corresponding to the excellent offspring, specifically comprising:

[0132] As shown in Figure 3 , inputting the offspring population Childgen into the trained SVM model to obtain offspring population labels

[0133] Calculating the number of labels greater than 0 in ChildTag gen and obtaining the corresponding index ChildIndex gen ={i1, i2,..., i k}, when the number is greater than or equal to SubCount, performing the next operation; when the number is less than SubCount, selecting individuals that perform well and whose predicted results are negative to make up, ChildIndex gen ={i1, i2,..., i k , j1,..., j l}, k+l==SubCount;

[0134] Obtaining the corresponding excellent offspring through the index ChildIndex gen of the excellent offspring:

[0135]

[0136] Performing a lower-level optimization process on the excellent offspring Child gen , when the excellent offspring Child​gen When all individuals in the population have completed the lower layer optimization process, the number of vehicles and the total length of the path corresponding to the excellent offspring are obtained.

[0137] The new generation population is selected, including the following steps:

[0138] The parent population p gen and the offspring population Child gen are merged into a temporary population Pop gen .

[0139] The size of the temporary population is recorded as PopCount, and the temporary population is non-dominantly sorted according to the two target values of the total length of the path and the number of vehicles, and the non-dominant sorting level and the crowded distance of each individual are calculated.

[0140] The crowded distance calculation formula is as follows:

[0141]

[0142] Where i is the ith individual in the temporary population, i < PopCount, M is the target number, j = 0 indicates that the target is the number of vehicles, j = 1 indicates that the target is the total length of the path traveled by the vehicle, and when the non-dominant level is the same, the crowded distance of the ith solution is the sum of the differences between the two adjacent solutions in each dimension target, and the crowded distance of the first and last individuals is set to infinity.

[0143] The temporary population is screened according to the non-dominant level and the crowded distance, and the individuals on the first front face are screened, when the screened first front face individuals exceed N u , N u individuals are screened out by the crowded distance, otherwise all individuals on the front face are selected; when the screened first front face individuals are less than N u , the remaining front face individuals are selected according to the rule until the selection is completed, and the new generation population P gen+1 is output.

[0144] The output new generation population P gen+1 is input into the SVM model, and the model is updated.

[0145] In step S5, the optimal route planning scheme is output, including the following steps:

[0146] According to the output new generation population, it is judged whether the current iteration generation gen is greater than the preset population generation Gen, and the preset population generation Gen of the present application is 10 generations, when the preset population generation Gen condition is not met, the iteration optimization process is continued; when the condition is met, the hyper volume HV value of the nearest Gen generation population is calculated, that is, The HV genThe maximum and minimum values in the above formula are denoted as hv max and hv min ;

[0147] The diversity change degree threshold of the population is 0.0001, the diversity change degree H of the population is calculated l When H l is greater than the diversity change degree threshold of the population (that is, when H l > 0.0001), the iterative optimization process is continued; otherwise, an individual with a dominance level IndividualLevel = 1 is randomly selected from the current population, the index of the individual is index, the optimal route scheme TaskOrder of the individual is obtained index and output, the number of vehicles and the total length of the path corresponding to the optimal path of the new generation population are the optimal route planning scheme:

[0148]

[0149] The application further provides a route planning system for a large-scale vehicle routing problem, comprising:

[0150] A task node acquisition module is configured to acquire a plurality of task nodes of the large-scale vehicle routing problem;

[0151] An upper layer clustering module is configured to group the plurality of task nodes, determine a clustering center of each group and a number of task node groups, and when a vehicle distribution task corresponding to a group to which the clustering center belongs meets a vehicle capacity limit requirement, distribute the task node to the corresponding group and form a corresponding population of an upper layer clustering scheme from each group;

[0152] A lower layer path optimization module is configured to determine a number of vehicles and a total length of a path corresponding to each individual in a population corresponding to each group, perform non-dominated sorting on each individual, determine a non-dominated sorting level of each individual as a label to construct a training sample, input the training sample into an SVM model, train the model, generate a child population by performing a difference mutation operation on a current population, input the child population into the trained SVM model, obtain a classification result of each individual in the child population, select excellent children and perform lower layer path optimization to obtain an optimal route corresponding to the excellent children, and merge the current population and the excellent children, perform non-dominated sorting on an upper layer to select a new generation population, and iteratively perform until the new generation population meets a preset termination condition, and output an optimal route planning scheme.

[0153] The route planning method for the large-scale vehicle routing problem provided by the application can well solve the large-scale vehicle routing problem with capacity constraints, effectively reduce the problem complexity, and thus obtain an optimal path implementation scheme.

[0154] Specifically, the upper layer of the double-layer optimization framework is used to find a more suitable clustering scheme by initializing different clustering schemes, so as to effectively overcome the problem of unsatisfactory clustering results caused by improper selection of initial centroids and clustering number. The lower layer is used to perform path optimization according to the clustering scheme presented by the upper layer, so as to obtain a corresponding optimal path, and the double-layer optimization framework is used to find a path planning scheme with shorter distance and lower cost.

[0155] The SVM model-based target domain determination method provided by the application filters out newly generated upper-layer clustering schemes by training an SVM model, wherein if a solution is non-dominated, the solution is marked as a positive class, otherwise the solution is marked as a negative class. Since path optimization is limited to promising upper-layer solutions, the reliability of path planning can be further enhanced.

[0156] The application determines the optimal path of the vehicle by combining the simulated annealing algorithm with local search, groups the task nodes according to the clustering scheme generated by the upper layer, and uses the simulated annealing algorithm to find the optimal path for each group. Since the simulated annealing algorithm is a global search algorithm, the generated solution is relatively random, and therefore it cannot be guaranteed that the "optimal route" obtained is the actual optimal route. The solution obtained above can be optimized by using a local search algorithm, and a better path planning scheme can be obtained.

[0157] Large-scale vehicle path optimization problem (VRP) is a key challenge in the logistics field, which requires efficient planning of the delivery routes of multiple vehicles. Reasonable optimization of the path can significantly reduce the cost, improve the delivery efficiency and service quality.

[0158] Embodiments

[0159] As shown in Figure 4 , the application provides the number of vehicles and the total length of the path corresponding to each parent in the obtained parents 1, 2 and 3, Figure 4 The 0 and 1 after the clustering number of each parent represent whether the i-th task node is selected as a clustering center. When the i-th task node is selected as a clustering center, the value is 1; otherwise, it is 0.

[0160] The offspring 1'-5' are generated according to the parents 1-3, and the classes of the offspring 1'-5' are predicted according to the trained SVM model. The offspring 1'-3' in the offspring 1'-5' are predicted as positive classes by the trained SVM model, and the offspring 4' and 5' are predicted as negative classes by the trained SVM model. The paths of the three offspring 1'-3' individuals with positive prediction results are screened out for lower-layer optimization, and the computing resources are allocated to obtain the optimal individual in the offspring 1'-3' as the offspring 2', and the corresponding total path length is 170. The offspring 4' and 5' with negative prediction results are discarded.

[0161] As shown in Figure 5As shown, the newly generated offspring 1'-3' and the parent 1-3 are combined to obtain a temporary population, the temporary population is non-dominantly sorted, and the vehicle number and the total path length are non-dominantly sorted as two targets to obtain a corresponding dominance level of each individual.

[0162] According to the obtained corresponding dominance level of each individual and the crowding distance, the vehicle number and the total path length are taken as comprehensive consideration factors, and the three individuals of the parent 1 and the offspring 2'-3' are comprehensively selected as a new generation population to perform a subsequent path optimization process.

[0163] Since the path optimization is limited to promising upper layer solutions, the embodiment can well solve the large-scale vehicle path problem with capacity constraints by filtering the upper layer clustering scheme, effectively reduce the problem complexity, and further strengthen the reliability of path planning.

[0164] The above describes only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

[0165] In addition, all technical and scientific terms used in the present application have the same meaning as generally understood by those skilled in the art of the present application, unless otherwise specified. All documents mentioned in the specification are incorporated by reference to disclose and describe the methods related to the documents. In the event of any conflict with any incorporated document, the content of the specification shall prevail.

Claims

1. A route planning method for a large scale vehicle routing problem, characterized in that, The method comprises the following steps: Obtaining a plurality of task nodes of a large-scale vehicle path; Grouping the plurality of task nodes, determining a cluster center of each group and a number of task node groups, and when a vehicle distribution task corresponding to a group to which the cluster center belongs meets a vehicle capacity limit requirement, assigning the task node to the corresponding group, and forming a corresponding population of an upper layer clustering scheme for each group; Determining a number of vehicles corresponding to each individual in the population of each group and a total length of a path, performing non-dominated sorting on each individual, determining a non-dominated sorting rank of each individual as a label to construct a training sample, inputting the training sample into an SVM model, and training the model; Generating a child population by performing a differential mutation operation on the current population; Inputting the child population into the trained SVM model to obtain a classification result of each individual in the child population, selecting an excellent child, performing lower layer path optimization, obtaining an optimal route corresponding to the excellent child, merging the current population and the excellent child, performing non-dominated sorting on the upper layer to select a new population, and optimizing the optimal route by using a local search algorithm, iteratively performing the optimization until the new population meets a preset termination condition, and outputting an optimal route planning scheme.

2. The route planning method of a large-scale vehicle routing problem according to claim 1, characterized in that, The determination of the cluster center of each group and the number of task node groups comprises the following steps: determining the current generation number , the first generation population: wherein, represents the generation of the number of clusters, represents the generation of the number of cluster centers, represents whether the generation of the individual selects the task node as a cluster center, represents that the task node is selected as a cluster center, represents that the task node is not selected as a cluster center, represents the population size; Initialization of population grouping information , represents the grouping information of the i-th individual of the j-th generation gen num th individual of the j-th generation​ According to Initializing individual task node groups , we get: wherein , g denotes the number of task node groups, and .

3. The route planning method of a large-scale vehicle routing problem according to claim 2, characterized in that, The assignment of the task node to the corresponding group and the formation of the corresponding population of the upper layer clustering scheme for each group comprises: Determine the minimum grouping number according to the ratio of task demand quantity and vehicle load Sort the tasks according to the demand quantity from large to small; According to the upper layer generated clustering scheme individual, grouping the task nodes, and preferentially assigning the task nodes with the largest demand amount during the task point assignment; Step 3.1: initializing num = 1; Step 3.2: Obtain the grouping information of the current individual ; Step 3.3: Initialization = 1; Step 3.4: Calculate task node distance to each cluster center, and determine to assign the task node to the group with the minimum distance j . After that, by judging whether the remaining capacity of the group j satisfies the capacity limit requirement of the warehouse, when satisfying the capacity limit requirement of the vehicle, the task node is assigned to the group j , and the record is recorded in ; otherwise, the task node is assigned to the cluster with the second minimum distance , and whether the remaining capacity of the group satisfies is judged, and step 3.4 is repeated until the assignment of the task node is completed and j = . Step 3.5: Take When L, execute the first The allocation process center of the task node, wherein L represents the number of distribution task nodes; when L, it proves that all task nodes of the num individual have been allocated to the group, and the group information of the current individual is updated by The group information of each task node recorded in the middle ; Step 3.6: Take num = num +1, when num is less than or equal to , execute Step 3.2; when num is greater than , it indicates that all individuals in the current population have completed the grouping process of the task node, and the iteration terminates.

4. The route planning method of a large-scale vehicle routing problem according to claim 3, characterized in that, The determination of the number of vehicles corresponding to each individual in the population of each group and the total length of the path comprises the following steps: Step 4.1: Initialization ; Step 4.2: Obtain the grouping information of the current individual ; Step 4.3: Initialization ; Step 4.4: execute the lower layer optimization process of the current group, and obtain the current group , , represent the number of task nodes of the current group, obtain the coordinates of the warehouse and the task nodes in this group, and seek the optimal driving path through the ant colony algorithm; the optimal driving path is locally optimized through the 2-opt algorithm to obtain the access order of the task nodes ={ }and the total length of the path of the task nodes ; Step 4.5: Update the total length of the path of the current task group and the order of node access of the current task group ; Step 4.6: Let , when , execute the lower-level optimization procedure for the j th group; when , it means that the individual's task grouping scheme has completed the lower-level optimization procedure, update the individual's path total length , the individual's number of vehicles , and the access order of the task node groups ; Step 4.7: Let num = Step 4.6 num + 1, and num if ≤ Step 4.2 is executed; otherwise, if num > Step 4.2 is not executed, and indicates that all individuals in the current population have completed the lower-level optimization process, and the iteration terminates.

5. The route planning method of the large-scale vehicle routing problem according to claim 4, characterized in that, The training of the model comprises: acquiring the total length of the path of each individual in the current population and the number of required vehicles ; determining a non-dominated rank of each individual according to the total length of the path of each individual and the number of vehicles wherein, denotes the non-dominated rank of the i-th individual in the j-th generation; 1≤i≤N ≤j ; and ; and ; and The individual label of the grade 1 is set to 1, and the individual label of other grades is set to -1 to obtain the label of the current population ; Obtaining a clustering solution for a population , represents a first generation of a first individual; The clustering scheme and labels are input into the SVM model, which is trained.

6. The route planning method of the large-scale vehicle routing problem according to claim 5, characterized in that, The generation of the child population comprises: Step 6.1 : Initialization of the offspring population ; Step 6.2: Initialization num = 1; Step 6.3: Get a cluster solution of the current individual , randomly select a non-dominated rank of the cluster solution of the individual of step 6.2 , randomly get an individual from the cluster solution of step 6.3 and ;​ Step 6.4: performing a differential operation to obtain an individual in the child population after the differential operation: wherein is a differential scaling factor; Step 6.5: Determine if the variable value in the offspring population after the differential operation exceeds the limits, and modify the variable value to the upper limit when greater than the upper limit, and to the lower limit when less than the lower limit. Step 6.5: Determine if the variable value in the offspring population after the differential operation exceeds the limits, and modify the variable value to the upper limit when greater than the upper limit, and to the lower limit when less than the lower limit. Step 6.6: Modify the variable after the polynomial mutation operation and check again whether the variable after the operation meets the requirement, and if not, modify the variable after the mutation operation according to step 6.

5. perform a polynomial mutation operation, and check again whether the variable after the operation meets the requirement, and if not, modify the variable after the mutation operation according to step 6.

5. Step 6.7: Let ,judge Whether it is true or false, if true, it means that the first... Offspring population Otherwise, proceed to step 6.

3.

7. The route planning method of the large-scale vehicle routing problem according to claim 6, characterized in that, The obtaining of the optimal route corresponding to the excellent child comprises: obtaining offspring population inputting the trained model, obtaining offspring population labels ; Computing the number of labels greater than 0 and obtaining the corresponding index When the number is greater than or equal to , the operation of obtaining excellent offspring is performed; when the number is less than , the individual with good performance whose prediction result is negative class is selected, ; By the index of the superior offspring Get the corresponding superior offspring: the elite offspring The lower-level optimization process is performed when all individuals in the elite offspring When all individuals in the elite offspring have completed the lower-level optimization process, the number of vehicles and the total length of the paths corresponding to the elite offspring are obtained.

8. The route planning method of the large-scale vehicle routing problem according to claim 7, characterized in that, The selection of the new population comprises the following steps: merging the parent population and the offspring population into a provisional population ; Let the size of the temporary population be denoted as The temporary population is ranked according to the two objective values of total path length and vehicle number, and the non-dominated rank and crowding distance of each individual are calculated. The crowded distance calculation formula is as follows: in, The first in the temporary population Individual, , For the target quantity, j =0 indicates that the target is the number of vehicles. j =1 indicates that the objective is the total length of the path traveled by the vehicle. When the non-dominance levels are the same, the first... The crowding distance of a solution is the sum of the differences in each dimension of the objective of the two adjacent solutions, where the crowding distance between the first individual and the last individual is set to infinity; According to the non-dominated rank and the crowding distance, the temporary population is screened, and the individuals of the first front are screened, when the screened individuals of the first front are more than , the individuals of are screened by the crowding distance, otherwise all the individuals on the front are selected; when the screened individuals of the first front are less than , the individuals of the remaining front are continuously selected according to the rule until the selection is finished, and the final population is output as the new generation population; output a new generation of offspring input to the model, updating the model.

9. The route planning method of the large-scale vehicle routing problem according to claim 8, characterized in that, The output of the optimal route planning scheme comprises the following steps: According to the new generation population , take , determine whether the current iteration number is greater than the preset population number Gen, when the preset population number Gen condition is not met, continue to execute the iterative optimization process; when the condition is met, calculate the hyper volume Gen value of the nearest HV generation population, that is , obtain the maximum value and the minimum value in and record them as and ; by calculating the degree of change in diversity of a new generation population : When If the diversity change degree of the current population is greater than the preset diversity change degree threshold, the iterative optimization process is continued; otherwise, an individual with a dominant rank is randomly selected from the current new population, and the index of the individual is The optimal route scheme of the individual is obtained, and the number of vehicles and the total length of the path corresponding to the optimal path of the new population are output as the optimal route planning scheme. ​​ 10. A route planning system for a large scale vehicle routing problem, characterized in that, Comprise: A task node acquisition module is configured to acquire a plurality of task nodes of a large-scale vehicle path; An upper layer clustering module is configured to group the plurality of task nodes, determine a cluster center of each group and a number of task node groups, and when a vehicle distribution task corresponding to a group to which the cluster center belongs meets a vehicle capacity limit requirement, assign the task node to the corresponding group, and form a corresponding population of an upper layer clustering scheme for each group; An upper layer clustering module is configured to group the plurality of task nodes, determine a cluster center of each group and a number of task node groups, and when a vehicle distribution task corresponding to a group to which the cluster center belongs meets a vehicle capacity limit requirement, assign the task node to the corresponding group, and form a corresponding population of an upper layer clustering scheme for each group; The lower-layer path optimization module is configured to determine the number of vehicles and the total length of paths corresponding to each individual in the population corresponding to each packet, perform non-dominated sorting on each individual, determine the non-dominated sorting rank of each individual, construct a training sample as a label, input the training sample into an SVM model, train the model, generate a child population by performing a differential mutation operation on the current population, input the child population into the trained SVM model, obtain a classification result of each individual in the child population, select excellent children and perform lower-layer path optimization to obtain an optimal route corresponding to the excellent children, merge the current population and the excellent children, perform non-dominated sorting at the upper layer, and select a new population; and perform optimization on the optimal route by using a local search algorithm, iteratively execute the optimization, and output an optimal route planning scheme when the new population meets a preset termination condition.

Images