CN112132312A - Path planning method based on evolution multi-objective multi-task optimization - Google Patents
Path planning method based on evolution multi-objective multi-task optimization Download PDFInfo
- Publication number
- CN112132312A CN112132312A CN202010818231.9A CN202010818231A CN112132312A CN 112132312 A CN112132312 A CN 112132312A CN 202010818231 A CN202010818231 A CN 202010818231A CN 112132312 A CN112132312 A CN 112132312A
- Authority
- CN
- China
- Prior art keywords
- population
- planning
- objective
- optimization
- updating
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
- G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/083—Shipping
- G06Q10/0835—Relationships between shipper or supplier and carriers
- G06Q10/08355—Routing methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- Bioinformatics & Computational Biology (AREA)
- Development Economics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Genetics & Genomics (AREA)
- Game Theory and Decision Science (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Physiology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a vehicle path planning method based on evolution multi-objective multi-task optimization, which takes each objective function of vehicle path planning as an independent task and solves a plurality of tasks simultaneously by utilizing operations of population structure, information migration, generation of filial generation, population updating and the like in the multi-task optimization; adopting a two-stage strategy of alternately performing multi-task optimization and multi-target optimization, switching the multi-task optimization process to the multi-target optimization process when a set switching condition is met, and optimizing a non-dominant solution set through operations of multi-target optimization such as population structure, generation of filial generation, population updating and external archive updating; and (3) adopting a population reconstruction strategy based on elite retention, retaining only part of elite solutions when set reconstruction conditions are met, and regenerating and adding the rest planning schemes into the population in a Gaussian walking-based mode. The method provided by the invention solves the problem that the optimization information of the similar problem can not be effectively exchanged and cooperated in the solving process, realizes the information sharing of the similar problem and improves the solving performance of VRP.
Description
Technical Field
The invention relates to the field of path planning, in particular to a path planning method based on evolution multi-objective multi-task optimization.
Background
The Vehicle Routing Problem (VRP) is a key link for optimization in modern logistics service, is also a link with the most service cost consumption in the logistics service, and is an indispensable support part for developing modern electronic commerce activities. On the other hand, postal delivery problems in the real world, airplane flight scheduling, railway vehicle marshalling, dock dispatching, water transportation ship dispatching, bus dispatching, power dispatching, medical resource dispatching in home care and the like can be abstracted as VRP. At present, most of research on vehicle path problems focuses on abstracting the vehicle path problems into a single-target or multi-target model, and then solving the vehicle path problems by using a single-target or multi-target optimization method.
In the research of the vehicle path problem, the target is generally solved separately, and the optimization information of similar problems cannot be effectively exchanged and cooperated in the solving process, so that the solving efficiency of the algorithm is limited.
Disclosure of Invention
The invention mainly aims to overcome the defects in the prior art, and provides a path planning method based on evolution multi-objective multi-task optimization, which adopts a solving paradigm of evolution multi-task optimization, solves the problem that optimization information of similar problems cannot be effectively exchanged and cooperated in the solving process, realizes information sharing of the similar problems, and improves the solving performance of VRP.
The invention adopts the following technical scheme:
a path planning method based on evolution multi-objective multi-task optimization specifically comprises the following steps:
s1: population initialization: generating an initial sequence set by adopting a random arrangement-based coding mode, then generating a planning scheme for each sequence in the set by adopting a construction method, evaluating according to an objective function, adding all the planning schemes into an initial population for updating an external archive, wherein the objective function is as follows:
f2(s)=maxk=1,..,Mc(rk)
wherein M represents the number of vehicle paths contained in the planning plan s, rkDenotes the kth vehicle path, c (r)k) Represents the distance traveled by the k-th vehicle route, f1(s) and f2(s) respectively represent the total distance traveled by the vehicle and the longest vehicle path length for the path planning scenario s.
S2: evolution multitask optimization: sequentially carrying out multitask population construction, multitask information migration, multitask offspring generation, multitask population updating and external archiving updating;
s3: two-stage switching: if the external archive is updated, the step S4 is executed to execute the multi-objective optimization phase; otherwise, returning to step S2, the multitask optimization phase continues to be executed.
S4: the evolution multi-objective optimization comprises the steps of sequentially carrying out multi-objective population structure, multi-objective offspring generation, multi-objective population update and external archive update;
s5: judging whether to execute reconstruction: if the algebra of the external archive which is not changed is larger than the set threshold value, the step S6 is entered, and the population reconstruction operation is executed; otherwise, the process proceeds to step S7.
S6: and (3) population reconstruction: and obtaining an optimal population according to a non-dominant sorting method and a crowding distance method, and forming a new population by adopting a Gaussian walking-based mode.
S7: judging whether the algorithm is finished: if the iteration times reach the set maximum iteration times, outputting all planning schemes in the external archive, and ending the algorithm; otherwise, return to step S2.
Specifically, the evolution multitask optimization in step S2: sequentially carrying out multitask population construction, multitask information migration, multitask offspring generation, multitask population updating and external archiving updating, and specifically comprising the following steps:
s21: multitask population construction: and calculating scalar adaptation values of all planning schemes of the initial population, then selecting the planning scheme with the largest scalar adaptation value, and constructing the population of the corresponding task.
S22: multi-task information migration: initializing a temporary population of the corresponding tasks, adding the planning scheme with the maximum scalar adaptation value in the population of the corresponding tasks constructed in the step 2 into the temporary population, and selecting the planning scheme with the maximum adaptation value from another task to add into the temporary population.
S23: and (3) generating a multitask filial generation: and executing the variation and cross operation of a differential evolution algorithm on the sequence corresponding to each planning scheme in the temporary population corresponding to the task, generating a filial generation sequence by using a maximum sequence value rule, obtaining the corresponding planning scheme and the objective function value of the planning scheme on the corresponding task by using a construction method, performing local search operation on each planning scheme, and replacing the original planning scheme in the temporary population with the obtained new planning scheme.
S24: multitask population updating and external archiving updating: and (3) calculating scalar adaptation values of all planning schemes in the population obtained in the step (2) and the temporary population obtained in the step (3) according to the objective function values, selecting the planning scheme with the largest scalar adaptation value to form a new population, and updating external archives of each planning scheme in the new population according to a non-dominated sorting and crowding distance method.
Specifically, the evolutionary multi-objective optimization in step S4 includes sequentially performing multi-objective population structure, multi-objective offspring generation, multi-objective population update, and external archive update, and specifically includes:
s41: multi-target population structure: and constructing a new multi-target population by using a non-dominated sorting and crowding distance selection method.
S42: generating multi-target filial generation: and performing sequential crossing operation and exchange variation operation on the multi-target population to generate new sequences, respectively constructing new planning schemes for the sequences by using a construction method, evaluating the new planning schemes, adding the non-dominated planning schemes into the corresponding multi-target temporary population, and performing local search operation.
S43: multi-target population updating and external archiving updating: and selecting a planning scheme from the multi-target population and the corresponding multi-target temporary population by adopting a non-dominated sorting and crowding distance method to form a new multi-target population, and simultaneously using the multi-target temporary population for updating external archives.
As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following advantages:
1. the invention takes each objective function of VRPRB as an independent task, solves a plurality of tasks simultaneously by utilizing operations of population structure, information migration, generation of filial generation, population updating and the like in multi-task optimization, adopts a two-stage strategy of alternately performing multi-task optimization and multi-objective optimization, switches the multi-task optimization process to the multi-objective optimization process when set switching conditions are met, accelerates the simultaneous solution of a plurality of vehicle path planning tasks by migrating similar information among different vehicle path planning tasks, solves the problem that the optimization information of the similar problem cannot effectively exchange and cooperate in the solving process, realizes the information sharing of the similar problem, and improves the solving performance of VRP.
2. Optimizing a non-dominant solution set by using an evolutionary multi-objective optimization strategy, namely, multi-objective optimized population structure, generation of filial generations, population updating, external archive updating and other operations; and guiding the generation process of a new planning scheme by a non-dominated sorting and crowding distance selection mode, and improving the diversity of the optimal planning scheme set.
3. And a population reconstruction strategy based on elite reservation is adopted, when a set reconstruction condition is met, only part of elite solutions are reserved, and other planning schemes are regenerated and added into the population in a Gaussian walking-based mode, so that the optimality of the planning schemes is further improved.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention.
The invention is described in further detail below with reference to the figures and specific examples.
Detailed Description
The invention is further described below by means of specific embodiments.
The VRP according to the present invention is a Vehicle Route Problem (VRPRB) with Route Balancing. The core of the vehicle path problem with path balancing, compared to other vehicle path problems, is to balance the workload of the various paths as much as possible while minimizing the total distance traveled. Therefore, the optimization goals of VRPRB are: total distance traveled and balance metrics. The first optimization objective of the vehicle path problem with path balancing is derived from the classical vehicle path problem and the second optimization objective is related to the balancing of the workload. For the second goal, there are currently several measurement methods: calculate the difference between the longest and shortest path lengths, minimize the longest path, and lexicographically minimum and maximum. According to the existing research results, the three different measurement methods have respective advantages and disadvantages in terms of balancing the workload. In the present invention, minimizing the longest path serves as a second optimization goal for the VRPRB model.
There are some adjustments to the order of steps in order to fully illustrate the content of this embodiment.
S101, problem model definition and solution representation
VRPRB can be defined as an optimization problem that includes two objectives, i.e. it is required to find an optimal set of vehicle path planning schemes to minimize the total driving distance while balancing the workload of each vehicle path as much as possible, and specifically, VRPRB is defined by a completely undirected graph G ═ N, E, where N ═ {0, 1.. N } is the node set and E { (i, j) | i, j ∈ N } is the edge set. The point set N consists of a yard 0 and N customers, and each node i belongs to N and has two attributes.
The solution for VRPRB (i.e. vehicle planning scheme) is defined as a set of paths consisting of M paths, s ═ r1,...,rMEach path can be represented asI.e. path rkIs composed of client points accessed in a back-and-forth order,denotes the parking lot, nkIs the number of client points visited in the k-th path. To ensure the effectiveness of the planning scenario s, it is required that each customer can only be included in one route and that the required capacity of the route cannot exceed the maximum capacity of the vehicle. Is provided withIs the distance traveled by the k-th route,indicating vehicle departure from customer locationTo the customer siteThe distance of travel of (a) is,representing the jth customer to be served on the kth path, the model of VRPRB can be expressed as:
min f={f1,f2}
wherein:
f2=maxk=1,..,Mc(rk)
constraint conditions are as follows:
first objective function f1Represents the planTotal distance travelled of the solution, second objective function f2Represents the maximum path length of the planning scenario, and C is the maximum capacity of the vehicle.
S102: construction of initial population
In this embodiment, an initial sequence is generated using a random permutation-based encoding scheme, each sequence representing a route from the yard (represented by the serial number "0"), sequentially visiting all customer site serial numbers, and returning to the yard. For each sequence, it is necessary to convert it into a feasible path planning scheme, depending on the number of vehicles. Here, two different target (or task) based construction methods will be used to segment a sequence into two different planning schemes. Generation of the sequence S ═ 0, π1,π2,...,πn) In which pijThe number of vehicles corresponding to the serial number of the jth visiting client is M. The first construction method (CM1) is based on the goal of total path length minimization. Specifically, L [ k ]][i]Indicating the use of m vehicles to visit a sequence of customer points (pi)1,π2,...,πn) The minimum driving distance of (a) is divided according to the following dynamic programming recursion formula:
L[k][i]=min0≤j≤i&&cap(j+1,i)≤C{L[m-1][j]+cost[j+1,i]}
wherein cos t (j, i) and cap (j, i) represent the sequence (0, πj,...,πi0) distance traveled and required capacity.
The second construction method (CM2) is designed based on the goal of minimizing the longest path length. F [ k ]][i]Representing m vehicles visiting a sequence of customer points (pi)1,π2,...,πn) The recursive formula of dynamic programming is as follows:
F[k][i]=min0≤j≤i&&cap(j+1,i)≤C{max{F[k-1][j],cost[j+1,i]}}
the above-described | OP | sequences are generated by a random permutation-based manner, and a construction method CM is applied to each sequence1And method of construction CM2Two planning schemes are generated respectively. Through this process, theGenerating 2 x OP plan schemes as initial population OPg. For each planning scenario, the evaluation is performed according to two objective functions, f1 and f 2.
S103, evolution multitask optimization strategy
In the evolutionary multitask optimization strategy, 4 components are mainly included: population construction, information migration, child generation and population updating. These 4 components will be described in detail below.
1) Multitask population structure
In VRPRB, each objective function is treated as an independent task. Therefore, in a multitask optimization environment, VRPRB can be defined as an optimization problem that includes two tasks w1 and w 1. The population construction process for task w1 is as follows:
first, OPgAccording to which all planning plans arei(sy) The values are sorted in a long sequence to obtain each planning scheme syFactor ranking r for task w1ij. Then, a scalar fitness value for the planning plan is calculated based on the factor rankingsFinally, h1 planning schemes with the largest scalar fitness value are selected as the group Pw1 of the task w 1.
2) Multitasking information migration
Information migration between different tasks is mainly achieved by migrating the best planning scheme of other tasks to the population of the task. In particular, the amount of the solvent to be used,
for task w1(w2), first, a temporary population is initialized for itThen, the h1 planning schemes with the maximum adaptive quantity values in the task group Pw1 are reserved and copied toIn (1). Then, h2 planning schemes with the largest scalar adaptation value are selected from tasks w2(w1), and are copied to TPw1(TPw 2).
3) Multitask child generation
For the sequences corresponding to each of the plans TPw1 and TPw2, new progeny sequences are generated by the mutation and crossover operations of the differential evolution algorithm. For the syThe planning scheme adopts the mutation operation and the cross operation which are respectively as follows:
vy=sbest+rand(0,1)*(sr1-sr2)
wherein v isyIndicates the variant solution generated, uj,lRepresents the resulting experimental solution ujThe l-dimension variable, sbestRepresents the best solution in the population, sr1,sr2Representing two different solutions randomly selected in the population, Cr representing the crossover rate. Since the resulting trial solution is not a sequence of integers, it needs to be converted into a sequence for the customer site. For this purpose, the conversion is carried out using the maximum order value rule: arranging each dimension of the test solution according to the numerical value from large to small; then, the dimension is set to the client point number corresponding to its rank.
Through this process, the sequence corresponding to each experimental solution can be obtained. Next, using the construction method CM1(CM2) The corresponding planning scheme and its objective function value at task w1(w2) are obtained.
To further improve the quality of the new planning scenario, for each planning scenario, sp is usedkTo perform a local search operation (section 5 will be described in detail).
Finally, the resulting new planning scheme is completely substituted TPw1(TPw2) for all planning schemes originally in it.
4) Multitask population update
In a multitasking environment, the population of each task is updated individually. For task w1, first, all planning solutions in Pw1(Pw2) and TPw1(TPw2) are sorted according to their objective function values on tasks w1 and w2, scalar fitness values thereof are calculated, and h1(h2) planning solutions with the largest scalar fitness values are selected to constitute a new Pw1(Pw 2). Meanwhile, each planning scenario in TPw1(TPw2) is used to update the external archive (section 7, described in detail).
S104 evolution multi-objective optimization strategy
The objective of the evolutionary multi-objective optimization strategy is to increase the diversity of the population through a non-dominated sorting and crowding distance selection method. The process of evolving a multi-objective optimization strategy mainly comprises 3 main components: population construction, child generation and population update. The method comprises the following specific steps:
1) multi-target population structure
In the evolutionary multi-objective optimization process, an algorithm sets an independent population for the evolutionary multi-objective optimization process. The construction method of the population is as follows: will OPg,Pw1And Pw2Combining a temporary population, selecting 2 × NP planning schemes from the temporary population by using a non-dominated sorting and crowding distance selection method in NSGA-II, and constructing a new multi-target population APg。
In the initial stage of the algorithm, i.e. when g is 1, let AP0=AP1As an initial population for a multi-objective optimization process.
2) Multi-target offspring generation
In embodiments of the present invention, sequential crossover operations and crossover mutation operations are employed to generate new progeny sequences. The specific operation is as follows:
and (3) sequential crossing operation: from the parent population (OP)g) Wherein the solution is selected for sequential interleaving. Firstly, randomly selecting a solution from a population; second, a second solution is selected, either randomly or based on path similarity. (the probability of using the path similarity selection method is set to 0.01). The path similarity of the two solutions is calculated based on the Jacccards similarity coefficient, i.e., the ratio of the number of shared arcs between the two solutions to the total number of arcs in the two solutions is calculated. In the population, each solution is associated with a similarity vector for storing the path similarity of the solution to the remaining solutions in the population. In the process of selecting solutions for crossover operation, when the first solutionAfter selection, the solution with the smallest path similarity value and not selected is selected as the second solution. Then, a sequential interleaving operation is performed on each set of solutions, resulting in two new solutions. The sequential interleaving operation is performed as follows: firstly, randomly selecting two segmentation points from one solution; secondly, copying the segment between the two segmentation points into the solution of the child thereof; then, deleting the elements appearing in the fragment in another solution; finally, the remaining elements in the other solution are inserted in order into the free positions of the first child solution. The other child solution is generated in a manner consistent with the first child solution. By sequential interleaving operations, | OP can be generatedgL new sequences.
Exchange mutation operation: for each newly generated child solution, crossover mutation operations are performed with a probability of 0.1. Specifically, two elements are randomly selected from the solution and their positions are exchanged, resulting in a new solution.
When the execution of the sequence crossing operation and the exchange mutation operation is completed, the newly generated | AP is processedgL sequences, respectively using CM1And CM2Construction 2. about. | APgAnd | planning schemes and evaluating. Then, adding the non-dominant planning scheme to the temporary population APg' of (1). Finally, the method of tournament selection is used to select from the APg∪APg' y planning plans are selected for local search operations (section 5, described in detail).
3) Multi-target population updates
AP is processed by adopting non-dominated sorting method and congestion distance calculation methodg∪APg' the combined population is sorted, and | AP is selected from the sorted combined populationgAnd l planning schemes form a new multi-target population. At the same time, the APg' for updating external archives.
Local search operation S105
In this embodiment, for the planning scheme, the following four common local search operations are adopted: such as:
exchange (j', j): path r1Client j's routing path r2Client j in (1)Instead, path r2Client j in (1) is represented by path r1Client j' of (1).
relocate (j', j): will route r1Client j' of (a) is removed and inserted into path r2Client j in the list.
2-opt (j', j): will route r1All customer points after customer j' are removed and inserted into path r2After client j in (1), then path r2All clients behind client j in (1) remove and insert path r1Of customer j'.
reverse (j', j): for a certain path in the scenario, the customer point between location j ' to location j is inverted so that this partial path changes from (j ', j ' +1 …, j-1, j) to (j, j-1 …, j ' +1, j ').
In the local searching process, firstly, a path with a longer driving distance is selected according to a roulette method; then, one of the above four local search operations is selected by probability to be performed. For each planning scheme, the above local search process Iter is repeatedly performed, Iter is the set number of local searches, and Iter in this embodiment is 30.
Selection probabilities sp for these four local search operationsk(k is 1,2,3, 4). Initially, the selection probabilities are all set to 1/4. At the end of each iteration, namely when the number of iterations reaches the set maximum value of the iteration number, updating the probability value of each local search operation in the following mode:
firstly, counting the selected times of each local search operation and the times of improving the quality of the original planning scheme, and respectively recording the times as ntkAnd nsk(k is 1,2,3, 4). The selection probability for each local search operation is then updated according to the following formula:
spk=(1-θ)*spk+θnsk/ntk
s106, population reconstruction mechanism
If Stagnation ≧ Grs holds (Stagnation denotes the number of generations with no changes to the external archive, Grs denotes a set threshold, then the population is considered to have entered the stall at this timeStatus. In order to solve the problem and further improve the searching performance of the algorithm, a population reconstruction mechanism is introduced. Specifically, the population OP is sorted according to the non-dominated sorting method and the method of the congestion distancegSorting is carried out, RS ═ α ×. 2 ×. NP are reserved for the best-ranked planning schemes, and α takes a value of 0-1. Other planning schemes will reinitialize by means of gaussian walk based approach, namely:
s′e RS=Gausslan(μ,σ)+(rand(0,1)*sRS+rand(0,1))
wherein s isRSIs a random one of RS planning schemes reserved, s'e RSFor the e-th reinitialized planning scheme, Gaussian (μ, σ) represents a Gaussian function with μ set to the selected sRSσ is set to
Reinitialized s'e RSInstead of a sequence of integers, it needs to be converted into a sequence for the client point. For this purpose, the conversion is carried out using the maximum order value rule: s'e RSEach dimension of (a) is arranged from large to small according to the value of the dimension; then, setting the dimension as the serial number of the client point corresponding to the ranking; then, using the construction method CM respectively1And CM2To two new planning scenarios.
The above reinitialization process (1- α) × 2 × NP times is repeated. And finally, sequencing the newly obtained 2 x (1-alpha) 2 x NP new planning schemes according to a non-dominant sequencing method and a crowding distance method, and selecting the best (1-alpha) 2 x NP planning schemes to be added into the population.
S107 external archive update mechanism
A dominance mechanism is employed when updating the external archive. Specifically, in the dominant archive, each non-dominant solution in the archive corresponds to one association vector B ═ B1,B2And (c) the step of (c) in which,
Bi=log(fi+1)/log(1+),
one non-dominated solution is stored in each hypercube. Thus, dominance-based archiving not only evenly distributes non-dominated solutions, but also effectively limits the size of external archives during the search.
The following specific test examples and experimental results are verified and explained;
to verify the effectiveness of the proposed method, the present invention uses seven CVRP reference examples proposed by Christofides and Eilon et al. In this example, the number of customers ranges from 50 to 199 people, and the number of vehicles is the minimum number of vehicles required to service these customers. The naming mode of the examples used in the invention refers to the mode used by Zhang et al, and the name of each example is as follows: ei-jk, where E indicates that the distances in the calculation are Euclidean distances; i is the number of vertices (including the yard). j is the number of available vehicles. k is a symbol indicating that the calculation is from Christofides or Eilon, if k ═ e, the calculation was by Christofides and Eilon et al, otherwise, the calculation was by Christofides et al. To obtain a fair experimental result, each example was run 10 times.
Based on the above test calculation set, the present invention selects the M-NSGA-II algorithm proposed by Zhang et al as the comparison (Zhang Z, Sun Y, Xie H, et al GMMA: GPU-based multi-objective metric algorithms for improving the resolution with the route balancing [ J ]. Applied Intelligence,2019,49(1): 63-78). The termination condition of the algorithm proposed by the present invention is set to maxGen 1000. To compare the two algorithms under the same conditions, the termination condition of the M-NSGA-II algorithm was also set to 1000. Statistics and analysis were performed by running 10 times each of the examples independently. In the case of minimizing the total travel distance, the proposed algorithm can obtain a smaller total travel distance than that of M-NSGA-II over 5 examples, respectively: E76-10E, E101-08E, E121-07c, E151-12c, and E200-17 c; the two algorithms perform equivalently on the examples E51-05E and E101-08E. In case of the minimized maximum path length, the proposed algorithm can obtain a smaller maximum path length than that of M-NSGA-II over 3 examples, respectively: E51-05E, E121-07c and E200-17 c; the algorithms proposed on examples E76-10E, E101-08E and E151-12c performed slightly worse. In general, the proposed algorithm is slightly better than M-NSGA-II, especially when optimizing total distance traveled.
In order to compare the performance of the proposed algorithm with that of the M-NSGA-II algorithm on seven reference examples, three indexes widely used by a multi-objective optimization algorithm are adopted, namely: IGD, HV and C-Metric. They are used to show the convergence and diversity of the algorithm from different perspectives, where IGD and HV are unitary indicators, and C-metric binary indicators. Aiming at the IGD index, the proposed algorithm is obviously superior to M-NSGA-II in 1 operator, has the same performance with M-NSGA-II in 6 operators, and does not have an operator which is obviously inferior to M-NSGA-II; aiming at HV indexes, the proposed algorithm is obviously superior to M-NSGA-II in 1 calculation example, has the same performance with M-NSGA-II in 6 calculation examples, and does not have calculation examples which are obviously inferior to M-NSGA-II; aiming at the C-Metric index, the proposed algorithm is remarkably superior to M-NSGA-II in 1 arithmetic example, has the same performance with M-NSGA-II in 5 arithmetic examples, and is remarkably inferior to M-NSGA-II in 1 arithmetic example. In general, the path planning method based on the evolutionary multi-objective multi-task optimization can provide a more effective solution frame for the vehicle path problem.
The above description is only an embodiment of the present invention, but the design concept of the present invention is not limited thereto, and any insubstantial modifications made by using the design concept should fall within the scope of infringing the present invention.
Claims (3)
1. A path planning method based on evolution multi-objective multi-task optimization specifically comprises the following steps:
s1: population initialization: generating an initial sequence set by adopting a random arrangement-based coding mode, then generating a planning scheme for each sequence in the set by adopting a construction method, evaluating according to an objective function, adding all the planning schemes into an initial population for updating an external archive, wherein the objective function is as follows:
f2(s)=maxk=1,..,M c(rk)
wherein M represents the number of vehicle paths contained in the planning plan s, rkDenotes the kth vehicle path, c (r)k) Represents the distance traveled by the k-th vehicle route, f1(s) and f2(s) respectively represent the total distance traveled by the vehicle and the longest vehicle path length for the path planning scenario s.
S2: evolution multitask optimization: sequentially carrying out multitask population construction, multitask information migration, multitask offspring generation, multitask population updating and external archiving updating;
s3: two-stage switching: if the external archive is updated, the step S4 is executed to execute the multi-objective optimization phase; otherwise, returning to step S2, the multitask optimization phase continues to be executed.
S4: the evolution multi-objective optimization comprises the steps of sequentially carrying out multi-objective population structure, multi-objective offspring generation, multi-objective population update and external archive update;
s5: judging whether to execute reconstruction: if the algebra of the external archive which is not changed is larger than the set threshold value, the step S6 is entered, and the population reconstruction operation is executed; otherwise, the process proceeds to step S7.
S6: and (3) population reconstruction: and obtaining an optimal population according to a non-dominant sorting method and a crowding distance method, and forming a new population by adopting a Gaussian walking-based mode.
S7: judging whether the algorithm is finished: the iteration times reach the set maximum iteration times, all planning schemes in an external archive are output, and the algorithm is ended; otherwise, return to step S2.
2. The method for path planning based on evolutionary multi-objective multi-task optimization of claim 1, wherein the evolutionary multi-task optimization in step S2: sequentially carrying out multitask population construction, multitask information migration, multitask offspring generation, multitask population updating and external archiving updating, and specifically comprising the following steps:
s21: multitask population construction: and calculating scalar adaptation values of all planning schemes of the initial population, then selecting the planning scheme with the largest scalar adaptation value, and constructing the population of the corresponding task.
S22: multi-task information migration: initializing a temporary population of the corresponding tasks, adding the planning scheme with the maximum scalar adaptation value in the population of the corresponding tasks constructed in the step 2 into the temporary population, and selecting the planning scheme with the maximum adaptation value from another task to add into the temporary population.
S23: and (3) generating a multitask filial generation: and executing the variation and cross operation of a differential evolution algorithm on the sequence corresponding to each planning scheme in the temporary population corresponding to the task, generating a filial generation sequence by using a maximum sequence value rule, obtaining the corresponding planning scheme and the objective function value of the planning scheme on the corresponding task by using a construction method, performing local search operation on each planning scheme, and replacing the original planning scheme in the temporary population with the obtained new planning scheme.
S24: multitask population updating and external archiving updating: and (3) calculating scalar adaptation values of all planning schemes in the population obtained in the step (2) and the temporary population obtained in the step (3) according to the objective function values, selecting the planning scheme with the largest scalar adaptation value to form a new population, and updating external archives of each planning scheme in the new population according to a non-dominated sorting and crowding distance method.
3. The path planning method based on evolutionary multi-objective multi-task optimization as claimed in claim 1, wherein the evolutionary multi-objective optimization in step S4 sequentially performs multi-objective population configuration, multi-objective offspring generation, multi-objective population update and external archive update, and specifically includes:
s41: multi-target population structure: and constructing a new multi-target population by using a non-dominated sorting and crowding distance selection method.
S42: generating multi-target filial generation: and performing sequential crossing operation and exchange variation operation on the multi-target population to generate new sequences, respectively constructing new planning schemes for the sequences by using a construction method, evaluating the new planning schemes, adding the non-dominated planning schemes into the corresponding multi-target temporary population, and performing local search operation.
S43: multi-target population updating and external archiving updating: and selecting a planning scheme from the multi-target population and the corresponding multi-target temporary population by adopting a non-dominated sorting and crowding distance method to form a new multi-target population, and simultaneously using the multi-target temporary population for updating external archives.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010818231.9A CN112132312B (en) | 2020-08-14 | 2020-08-14 | Path planning method based on evolutionary multi-objective multi-task optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010818231.9A CN112132312B (en) | 2020-08-14 | 2020-08-14 | Path planning method based on evolutionary multi-objective multi-task optimization |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112132312A true CN112132312A (en) | 2020-12-25 |
CN112132312B CN112132312B (en) | 2022-08-23 |
Family
ID=73851600
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010818231.9A Active CN112132312B (en) | 2020-08-14 | 2020-08-14 | Path planning method based on evolutionary multi-objective multi-task optimization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112132312B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112819255A (en) * | 2021-03-08 | 2021-05-18 | 吉林大学 | Particle swarm-genetic algorithm-based multi-criterion ship route determining method and device, computer equipment and storage medium |
CN112884229A (en) * | 2021-02-26 | 2021-06-01 | 中新国际联合研究院 | Large-scale public place people flow guiding path planning method based on differential evolution algorithm |
CN113050422A (en) * | 2021-03-09 | 2021-06-29 | 东北大学 | Multi-robot scheduling method based on maximin function multi-objective optimization algorithm |
CN113792989A (en) * | 2021-08-24 | 2021-12-14 | 武汉理工大学 | Demand-driven shared sightseeing vehicle inter-area parallel optimization scheduling method |
CN114510072A (en) * | 2022-01-18 | 2022-05-17 | 香港理工大学深圳研究院 | Multi-unmanned aerial vehicle path planning method, terminal and medium based on evolution migration optimization |
CN115574826A (en) * | 2022-12-08 | 2023-01-06 | 南开大学 | National park unmanned aerial vehicle patrol path optimization method based on reinforcement learning |
Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002187622A (en) * | 2000-12-21 | 2002-07-02 | Honda Express Co Ltd | Method of optimizing loading pattern in vehicle transportation system and base server for the same and memory medium recorded with program for the same |
US20160033292A1 (en) * | 2014-07-30 | 2016-02-04 | International Business Machines Corporation | Method and system for obtaining trajectory pattern of route |
CN105894222A (en) * | 2014-12-16 | 2016-08-24 | 重庆邮电大学 | Logistics distribution path optimization method |
CN107506846A (en) * | 2017-07-10 | 2017-12-22 | 北京石油化工学院 | A kind of vehicle dispatching method and device based on multi-objective particle |
CN108182499A (en) * | 2018-01-25 | 2018-06-19 | 上海交通大学 | A kind of hybrid ant colony for VRP problems and its realize system |
WO2018220439A2 (en) * | 2017-05-30 | 2018-12-06 | Nauto Global Limited | Systems and methods for safe route determination |
CN109635998A (en) * | 2018-11-02 | 2019-04-16 | 华侨大学 | A kind of adaptive Multipurpose Optimal Method solving vehicle routing problem with time windows |
CN109764882A (en) * | 2018-12-27 | 2019-05-17 | 华侨大学 | A kind of multiple target vehicle path planning method based on adaptive local search chain |
CN110516871A (en) * | 2019-08-22 | 2019-11-29 | 安庆师范大学 | A kind of dynamic vehicle method for optimizing route based on fuzzy roll stablized loop strategy |
CN110569959A (en) * | 2019-09-05 | 2019-12-13 | 南宁师范大学 | Multi-target particle swarm optimization algorithm based on collaborative variation method |
CN110598863A (en) * | 2019-09-05 | 2019-12-20 | 南宁师范大学 | Multi-target differential evolution method for co-evolution |
CN110766211A (en) * | 2019-10-14 | 2020-02-07 | 中国地质大学(武汉) | Method for creating vehicle path planning problem model under real-time road condition |
US20200050209A1 (en) * | 2018-08-07 | 2020-02-13 | GM Global Technology Operations LLC | Intelligent vehicle navigation systems, methods, and control logic for deriving road segment speed limits |
CN111144568A (en) * | 2019-12-19 | 2020-05-12 | 华南理工大学 | Multi-target urban logistics distribution path planning method |
CN111178485A (en) * | 2019-09-05 | 2020-05-19 | 南宁师范大学 | Multi-target evolutionary algorithm based on double population cooperation |
CN111445094A (en) * | 2020-04-28 | 2020-07-24 | 宁德师范学院 | Express vehicle path optimization method and system based on time requirement |
JP6739078B1 (en) * | 2020-01-14 | 2020-08-12 | 三東運輸株式会社 | Route calculation program, route optimization system, and route calculation method |
-
2020
- 2020-08-14 CN CN202010818231.9A patent/CN112132312B/en active Active
Patent Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002187622A (en) * | 2000-12-21 | 2002-07-02 | Honda Express Co Ltd | Method of optimizing loading pattern in vehicle transportation system and base server for the same and memory medium recorded with program for the same |
US20160033292A1 (en) * | 2014-07-30 | 2016-02-04 | International Business Machines Corporation | Method and system for obtaining trajectory pattern of route |
CN105894222A (en) * | 2014-12-16 | 2016-08-24 | 重庆邮电大学 | Logistics distribution path optimization method |
WO2018220439A2 (en) * | 2017-05-30 | 2018-12-06 | Nauto Global Limited | Systems and methods for safe route determination |
CN107506846A (en) * | 2017-07-10 | 2017-12-22 | 北京石油化工学院 | A kind of vehicle dispatching method and device based on multi-objective particle |
CN108182499A (en) * | 2018-01-25 | 2018-06-19 | 上海交通大学 | A kind of hybrid ant colony for VRP problems and its realize system |
US20200050209A1 (en) * | 2018-08-07 | 2020-02-13 | GM Global Technology Operations LLC | Intelligent vehicle navigation systems, methods, and control logic for deriving road segment speed limits |
CN109635998A (en) * | 2018-11-02 | 2019-04-16 | 华侨大学 | A kind of adaptive Multipurpose Optimal Method solving vehicle routing problem with time windows |
CN109764882A (en) * | 2018-12-27 | 2019-05-17 | 华侨大学 | A kind of multiple target vehicle path planning method based on adaptive local search chain |
CN110516871A (en) * | 2019-08-22 | 2019-11-29 | 安庆师范大学 | A kind of dynamic vehicle method for optimizing route based on fuzzy roll stablized loop strategy |
CN110569959A (en) * | 2019-09-05 | 2019-12-13 | 南宁师范大学 | Multi-target particle swarm optimization algorithm based on collaborative variation method |
CN110598863A (en) * | 2019-09-05 | 2019-12-20 | 南宁师范大学 | Multi-target differential evolution method for co-evolution |
CN111178485A (en) * | 2019-09-05 | 2020-05-19 | 南宁师范大学 | Multi-target evolutionary algorithm based on double population cooperation |
CN110766211A (en) * | 2019-10-14 | 2020-02-07 | 中国地质大学(武汉) | Method for creating vehicle path planning problem model under real-time road condition |
CN111144568A (en) * | 2019-12-19 | 2020-05-12 | 华南理工大学 | Multi-target urban logistics distribution path planning method |
JP6739078B1 (en) * | 2020-01-14 | 2020-08-12 | 三東運輸株式会社 | Route calculation program, route optimization system, and route calculation method |
CN111445094A (en) * | 2020-04-28 | 2020-07-24 | 宁德师范学院 | Express vehicle path optimization method and system based on time requirement |
Non-Patent Citations (4)
Title |
---|
DEMING PENG ETC: ""Experimental Analysis of Selective Imitation for Multifactorial Differential Evolution"", 《SPRINGERLINK》 * |
KAI KAI ZHANG ETC: ""Multiobjective memetic algorithm based on adaptive local search chains for vehicle routing problem with time windows"", 《SPRINGERLINK》 * |
YIQIAO CAI ETC: ""Differential evolution with hybrid linkage crossover"", 《ELSEVIER》 * |
YIQIAO CAI ETC: ""Differential Evolution With Neighborhood and Direction Information for Numerical Optimization"", 《IEEE》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112884229A (en) * | 2021-02-26 | 2021-06-01 | 中新国际联合研究院 | Large-scale public place people flow guiding path planning method based on differential evolution algorithm |
CN112819255A (en) * | 2021-03-08 | 2021-05-18 | 吉林大学 | Particle swarm-genetic algorithm-based multi-criterion ship route determining method and device, computer equipment and storage medium |
CN113050422A (en) * | 2021-03-09 | 2021-06-29 | 东北大学 | Multi-robot scheduling method based on maximin function multi-objective optimization algorithm |
CN113792989A (en) * | 2021-08-24 | 2021-12-14 | 武汉理工大学 | Demand-driven shared sightseeing vehicle inter-area parallel optimization scheduling method |
CN113792989B (en) * | 2021-08-24 | 2024-01-30 | 武汉理工大学 | Demand-driven parallel optimization scheduling method between shared sightseeing vehicle areas |
CN114510072A (en) * | 2022-01-18 | 2022-05-17 | 香港理工大学深圳研究院 | Multi-unmanned aerial vehicle path planning method, terminal and medium based on evolution migration optimization |
CN114510072B (en) * | 2022-01-18 | 2022-12-06 | 香港理工大学深圳研究院 | Multi-unmanned aerial vehicle path planning method, terminal and medium based on evolution migration optimization |
CN115574826A (en) * | 2022-12-08 | 2023-01-06 | 南开大学 | National park unmanned aerial vehicle patrol path optimization method based on reinforcement learning |
Also Published As
Publication number | Publication date |
---|---|
CN112132312B (en) | 2022-08-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112132312B (en) | Path planning method based on evolutionary multi-objective multi-task optimization | |
Tom et al. | Transit route network design using frequency coded genetic algorithm | |
Ribeiro et al. | Ant colony optimization: an overview | |
CN111178582B (en) | Logistics distribution optimization method based on improved genetic algorithm | |
CN111144568A (en) | Multi-target urban logistics distribution path planning method | |
CN112013829B (en) | Multi-UAV/UGV collaborative long-term operation path planning method based on multi-objective optimization | |
Mauttone et al. | A multi-objective metaheuristic approach for the transit network design problem | |
Duman et al. | Branch-and-price-and-cut methods for the electric vehicle routing problem with time windows | |
CN109764882A (en) | A kind of multiple target vehicle path planning method based on adaptive local search chain | |
CN111461402B (en) | Logistics scheduling optimization method and device, computer-readable storage medium and terminal | |
CN115576343B (en) | Multi-target vehicle path optimization method combined with unmanned aerial vehicle distribution | |
CN110098964A (en) | A kind of disposition optimization method based on ant group algorithm | |
Ashouri et al. | A Combination of Meta-heuristic and Heuristic Algorithms for the VRP, OVRP and VRP with Simultaneous Pickup and Delivery | |
Maleki et al. | A hybrid algorithm for the open vehicle routing problem | |
CN113848970A (en) | Multi-target collaborative path planning method for vehicle and unmanned aerial vehicle | |
Pylyavskyy et al. | A reinforcement learning hyper-heuristic for the optimisation of flight connections | |
CN115983755A (en) | Multi-type combined transport path optimization method | |
Lan et al. | Region-focused memetic algorithms with smart initialization for real-world large-scale waste collection problems | |
CN108108554B (en) | Multi-material vehicle body assembly sequence planning and optimizing method | |
Goel et al. | Improved multi-ant-colony algorithm for solving multi-objective vehicle routing problems | |
Tan et al. | Multi-objective teaching–learning-based optimization algorithm for carbon-efficient integrated scheduling of distributed production and distribution considering shared transportation resource | |
CN114415690A (en) | Multi-agent path planning method, navigation server and readable storage medium | |
Alaia et al. | A Comparative Study of the PSO and GA for the m-MDPDPTW | |
Derigs et al. | Indirect search for the vehicle routing problem with pickup and delivery and time windows | |
Zhou et al. | MVMoE: Multi-Task Vehicle Routing Solver with Mixture-of-Experts |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB03 | Change of inventor or designer information |
Inventor after: Cai Yiqiao Inventor after: Fu Shunkai Inventor after: Wang Yufei Inventor after: Zeng Shengming Inventor after: Liu Hongzhao Inventor before: Cai Yiqiao Inventor before: Fu Shunkai Inventor before: Zeng Shengming Inventor before: Liu Hongzhao |
|
CB03 | Change of inventor or designer information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |