CN108596469A - A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems - Google Patents

A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems Download PDF

Info

Publication number
CN108596469A
CN108596469A CN201810355489.2A CN201810355489A CN108596469A CN 108596469 A CN108596469 A CN 108596469A CN 201810355489 A CN201810355489 A CN 201810355489A CN 108596469 A CN108596469 A CN 108596469A
Authority
CN
China
Prior art keywords
strategy
strategy combination
extensive
iteration
quick self
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
Application number
CN201810355489.2A
Other languages
Chinese (zh)
Other versions
CN108596469B (en
Inventor
阳旺
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN201810355489.2A priority Critical patent/CN108596469B/en
Publication of CN108596469A publication Critical patent/CN108596469A/en
Application granted granted Critical
Publication of CN108596469B publication Critical patent/CN108596469B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063Operations research, analysis or management
    • G06Q10/0637Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
    • G06Q10/06375Prediction of business process outcome or impact based on a proposed change

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Educational Administration (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Navigation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems that the invention discloses a kind of, by adding period scoring and strategy combination weight regulation mechanism, using n times iteration as a cycle, the performance quality situation in the time, which is executed, according to each strategy combination unit in the period carries out weight adjustment, increase the selected probability of the strategy combination by improving the weight of the strategy combination to do very well, reduces the weight of the strategy combination of performance difference to reduce the selected probability of the strategy combination.With going deep into for iteration, unit, which executes performance superior strategy combination in the time, has the probability of bigger selected, realizes quick self-adapted.

Description

A kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems Method
Technical field
The quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems that the present invention relates to a kind of.
Background technology
Vehicle Routing Problems are commonly defined as:Several vehicles are from one or more central points, to there is different cargos Several customer's points of demand provide service, and roadway appropriate is planned in the case where meeting customer's point demand and certain constraints Line achievees the purpose that transportation cost is minimum.
Vehicle Routing Problems mainly have exact algorithm and heuritic approach two major classes method for solving.For the vehicle of small-scale Routing problem, the exact algorithms such as branch and bound algorithms, dynamic programming algorithm can be solved successfully, but Vehicle Routing Problems The time is solved as problem scale increase exponentially increases, this makes exact algorithm be difficult to solve large-scale vehicle route to ask Topic.Then, people begin one's study heuritic approach, and wherein savings method, scanning method, two-phase method etc. is most representative.It is close Year, heuritic approach is fast-developing, genetic algorithm, simulated annealing, tabu search algorithm, large neighborhood search algorithm etc. Modern optimization algorithm is also applied in the solution of Vehicle Routing Problems.
Large neighborhood search algorithm is the extension for destroying algorithm for reconstructing, and main thought is to rebuild principle according to destruction, Several are built to destroy the factor (table 3) and rebuild the factor (table 4).It destroys the factor and the reconstruction factor is combined to be formed to destroy two-by-two and be rebuild Strategy combination (table 1) sets certain weight in algorithm initialization for each strategy combination, in each step iterative process, Iteration object is destroyed and is rebuild according to a kind of strategy combination of certain policy selection, generate new explanation, according to acceptance criterion come Judge whether to receive the new explanation, and preserve current optimal solution, iterative cycles export optimal solution after meeting stopping criterion for iteration. In iterative process, each strategy combination has certain probability selected, has the search of bigger empty compared with destroying algorithm for reconstructing Between, it is absorbed in the possibility smaller of local optimum, increases the possibility for obtaining globally optimal solution, so being referred to as extensive field Searching algorithm.In iterative process each time, the selected probability of each strategy combination and its weight are closely bound up, but big In scale Neighborhood-region-search algorithm, the weight of each strategy combination is just had determined during algorithm initialization and algorithmic procedure In will not change.This results in the flexibility of algorithm relatively low, causes to solve that the time is long, and the undesirable technology of effect solved is asked Topic.
Invention content
The purpose of the present invention is in large neighborhood search algorithm (Large Neighborhood Search Algorithm, LNS) on the basis of a kind of quickly adaptive large neighborhood search algorithm (Fast Adaptive are provided Large Neighborhood Search Algorithm, FALNS) it solves in enterprise practical logistics distribution process due to dispatching Scale increases the problem of original traditional algorithm can not provide a high quality feasible solution in a short time.
In order to achieve the above technical purposes, the technical scheme is that,
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, including following step Suddenly:
Step 1:Parameter training, based on historical data to period scoring machine in quick self-adapted large neighborhood search algorithm The involved grading parameters of system are trained;
Step 2:The iterations needed for a cycle are completed in initialization, setting, are given birth to using RegretInsert insertions At initial solution;
Step 3:From the combination of multiple destruction Reconstruction Strategies a strategy combination is selected according to wheel disc method;
Step 4:A solution is randomly choosed from current several optimal solution sets Solutions is used as iteration object currentSolution;
Step 5:The destruction Reconstruction Strategy selected in applying step 3 combines the iteration object to being selected in step 4 CurrentSolution carries out destruction reconstruction, generates new explanation newSolution, if newSolution is better than current optimal solution BestEver then updates bestEver with newSolution;
Step 6:Determined whether to receive new explanation newSolution according to threshold value acceptance criterion, updates optimal solution if receiving Set Solutions;
Step 7:Judge whether to meet algorithm iteration end condition, step 10 is passed directly to if meeting;Otherwise judge to work as Whether the preceding period terminates, and step 3 is gone to if being not over, if it is no then follow the steps 8 and be not over go to step 3.Wherein Stopping criterion for iteration can be set as meeting preset maximum iteration, meet improvement ratio of solution etc..If being unsatisfactory for iteration end Only condition then continues iteration.Before continuing iteration, first judge whether current period terminates, i.e., whether completes Iterations in preset a cycle, to carry out the scoring of strategy combination to current period.
Step 8:It is scored each strategy combination according to period scoring.It is current optimal that generation is rebuild in primary destruction Solution, corresponding strategy combination plus s1Point;Primary destroy is rebuild without generating current optimal solution, but is improved than iteration object, accordingly Strategy combination adds s2Point;It is primary to destroy the new explanation rebuild and generated than iteration to aberration, but received by acceptance criterion, corresponding plan Slightly combination plus s3Point.The overall score of each strategy combination in measurement period calculates each strategy combination unit and executes in the time Score;
Step 9:Each strategy combination history weight and each strategy in the new round period are combined according to weight update mechanism Composite unit executes the score in the time and carries out weight adjustment, then goes to step 3;
Step 10:Export optimal solution bestEver.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described In step 1, the grading parameters include s1、s2And s3, be respectively used in quick self-adapted large neighborhood search algorithm into Row is primary to destroy the three kinds of different new explanation situation C generated after reconstruction1、C2And C3It scores, wherein C1Indicate that this time destroys weight It builds and generates current optimal solution, corresponding strategy combination plus s1Point, C2It indicates that this time destroys to rebuild without generating current optimal solution, but compares Iteration object is improved, corresponding strategy combination plus s2Point, C3Indicate that this time destroys the new explanation rebuild and generated than iteration to aberration, but It is to be received by acceptance criterion, corresponding strategy combination plus s3Point.If generate new explanation do not received by acceptance criterion, not into Row scoring.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described In step 2, initial solution is generated using RegretInsert insertions, is to calculate its sub-optimal insertion position for each ingress to be inserted With the cost difference of best insertion position, the maximum node of cost difference is inserted into its best insertion position.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described In step 3, the destruction Reconstruction Strategy is from including RandomRuin, RadialRuin, WorstRuin, ClusterRuin Destruction algorithm in choose any one kind of them as destroy the factor, while from including BestInsert, RegretInsert, Chosen any one kind of them in the algorithm for reconstructing of GreedyInsert as the factor is rebuild, be combined and the destruction Reconstruction Strategy that is formed.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described In step 3, according to wheel disc method select a strategy combination, be by following formula come use tendency in selection Great possibility Wheel disc method carry out calculative strategy and combine the probability that is chosen to:
Wherein k is the total number of strategy combination, πiIndicate the scoring of i-th kind of strategy combination.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described In step 4, is stored in the optimal solution set Solutions and be not more than a current optimal solutions of M.Wherein M is one pre- If parameter, i.e., directly take scoring it is highest it is preceding M solution conduct current optimal solution set.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described In step 5, the particular content combined by the destruction Reconstruction Strategy selected in step 3 carries out a part of node currently solved It destroys, this part of nodes is removed, then generates new explanation newSolution by repairing this part.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described Step 6 includes following procedure:
In an iterative process, if current solutions set elements are less than M, directly receive newSolution simultaneously Update solutions set;Otherwise newSolution and iteration object currentSolution are compared, according to threshold value Acceptance criterion updates solutions set;
The threshold value acceptance criterion is to receive the range of poor solution by continuously decreasing and tend to global optimum, and threshold value connects It is as follows by criterion formulas:
costnew< costcurrent+currentThreshold
Wherein costnewFor the expense of original solution, costcurrentFor the expense currently connect;
Present threshold value currentThreshold calculation formula are:
Initial threshold initialThreshold calculation formula are:
InitThreshold=costInitial solution*THRESHOLD_INI
Wherein i is current iteration number, costInitial solutionIt is the expense of initial solution, maximum iteration maxIterations, Initial threshold coefficient T HRESHOLD_INI and changes of threshold coefficient alpha are set in algorithm initialization.
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described When step 8 executes, within each period, the corresponding score value scoreTotal of each strategy combination need to be countediExist with each strategy Execution time exeTime in the period in totali, and calculate in each period institute's band in each strategy combination executable unit time The score addition score comei
A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, it is described Weight update mechanism in step 9 is by for caused point in each strategy combination executable unit time in each period Number addition scoreiThe weight of strategy combination is adjusted:
Wherein, ∑ scoreiAfter for current period, caused score in each strategy combination executable unit time Addition scoreiThe sum of, r is the history significance level factor, controls the speed of weight variation.R is not less than 0 and is not more than 1.Work as r When being 1, weight does not change, has reformed into large neighborhood search algorithm;When r is 0, new round weight only with work as The corresponding score value of each strategy combination is related in the preceding period, is shown experimentally that, r values (0.9,0.95) are more suitable.wijIt is The weight of the upper each strategy combination of a cycle, i.e. historical factor.
The technical effects of the invention are that the present invention compared to large neighborhood search algorithm for, can be vehicle road While diameter problem provides high quality solution, additionally it is possible to substantially shorten than large neighborhood search algorithm and solve the time, be more suitable for big Scale Vehicle Routing Problems.
The invention will be further described below in conjunction with the accompanying drawings.
Description of the drawings
Fig. 1 is flow chart of the present invention;
Fig. 2 is to destroy to rebuild principle schematic;
Fig. 3 is that each destruction Reconstruction Strategy of individual data collection of the present invention combines change curve schematic diagram;
Fig. 4 is that the single Reconstruction Strategy that destroys of the present invention combines the change curve schematic diagram concentrated in different data.
Specific implementation mode
Quickly adaptive large neighborhood search algorithm proposed by the present invention is the extension of large neighborhood search algorithm. In large neighborhood search algorithm, the weight of each strategy combination will not adjust after algorithm initialization, adaptive extensive Neighborhood-region-search algorithm adds period scoring and strategy combination weight regulation mechanism.Using n times iteration as a cycle, this reality It is 200 to apply N values in example, and executing the performance quality situation in the time according to each strategy combination unit in the period carries out weight tune It is whole, increase the selected probability of the strategy combination by improving the weight of the strategy combination to do very well, reduces the plan of performance difference The weight slightly combined reduces the selected probability of the strategy combination.With going deep into for iteration, unit execute in the time performance compared with Good strategy combination has the probability of bigger selected, realizes quick self-adapted.
Referring to Fig. 1, the present invention includes the following steps:
Step 1:Parameter training, based on large neighborhood search algorithm using relevant historical data to quick self-adapted big rule Grading parameters in mould Neighborhood-region-search algorithm involved by period scoring are trained;
Step 2:Initialization generates initial solution using RegretInsert insertions;
Step 3:From the combination of multiple destruction Reconstruction Strategies a strategy combination is selected according to wheel disc method;It is specific to destroy weight It builds strategy combination and can be found in the following table 1:
1 strategy combination table of table
Given in upper table partial strategy combination example, actually destroy Reconstruction Strategy be from including RandomRuin, Choose any one kind of them in the destruction algorithm of RadialRuin, WorstRuin, ClusterRuin as destroy the factor, while from including It is chosen any one kind of them in the algorithm for reconstructing of BestInsert, RegretInsert, GreedyInsert as the factor is rebuild, i.e., it can also shape At more different strategy combinations, do not expressed all in table.
Step 4:A solution is randomly choosed from current several optimal solution sets Solutions is used as iteration object currentSolution;
Step 5:Referring to Fig. 2, the destruction Reconstruction Strategy selected in applying step 3 combines the iteration pair to being selected in step 4 As currentSolution carries out destruction reconstruction, new explanation newSolution is generated, and judge whether to update current optimal solution bestEver;It wherein commonly destroys phase algorithm and can be found in the following table 2:
2 score by rules table of table
Common phase of regeneration algorithm can be found in the following table 3:
Table 3 often uses phase of regeneration algorithm
Step 6:Determined whether to receive new explanation newSolution according to threshold value acceptance criterion, updates optimal solution if receiving Set Solutions;
Step 7:Judge whether to meet algorithm iteration end condition, step 10 is passed directly to if meeting, otherwise judges to work as Whether the preceding period terminates, and step 3 is gone to if being not over, no to then follow the steps 8.Algorithm iteration in the present embodiment terminates item Part is set as meeting preset iterations, and specific number is 50000.
Step 8:It is scored each strategy combination according to period scoring, it is current optimal that generation is rebuild in primary destruction Solution, corresponding strategy combination plus s1Point;Primary destroy is rebuild without generating current optimal solution, but is improved than iteration object, accordingly Strategy combination adds s2Point;It is primary to destroy the new explanation rebuild and generated than iteration to aberration, but received by acceptance criterion, corresponding plan Slightly combination plus s3Point.The overall score of each strategy combination in measurement period calculates each strategy combination unit and executes in the time Score;Score by rules is referring to the following table 4:
4 score by rules table of table
Step 9:Each strategy combination history weight and each strategy in the new round period are combined according to weight update mechanism Composite unit executes the score in the time and carries out weight adjustment, then goes to step 3;
Step 10:Export optimal solution bestEver.
1) training parameter
It is that iteration scores each time in the period present invention introduces period scoring, so needing to determine s in advance1、 s2、s3Parameter value.The present invention is based on large neighborhood search algorithm using historical data to s1、s2、s3Parameter value is instructed Practice.15 Solomon data sets of the following table 5 pair are tested, the C in each data set counts 10000 iteration respectively1、C2、C3 The number that this 3 kinds of situations occur, the complexity occurred according to 3 kinds of situations is to s1、s2、s3Carry out value:
5 three kinds of situation occurrence number comparisons of table
Table 5 is with C1Based on the number of appearance, C has been calculated separately in each data set2、C3The number that two kinds of situations occur With C1Ratio, then remove respectively after highest and lowest two amount to 4 ratios, it is flat according to remaining 11 ratio calculations Obtain a general ratio.It can be seen that C3The number of appearance is probably C125 times, C2The number of appearance is probably C15 times.The complexity that 3 kinds of situations occur should be inversely proportional with corresponding score value, so if s3It is 1, then s2And s3Point It is not 5 and 25.The data set of training parameter and the data set of subsequent experimental are identical herein, have higher coupling to a certain extent It is right.But in enterprise practical case, parameter appropriate can be trained using historical data, as daily routine data The foundation of processing, so having certain realistic meaning.
2) it initializes
The present invention generates initial solution using RegretInsert algorithms.The generating process of initial solution is an iterative process, Each iteration finds a node and carries out insertion operation.RegretInsert algorithms are each be inserted into each iterative process Node calculates the cost difference of its sub-optimal insertion position and best insertion position, and the maximum node of cost difference is inserted into it most (if only best insertion position, without sub-optimal insertion position, then sub-optimal insertion position cost is arranged to very big for good insertion position An integer), in next iterative process, it is only necessary to recalculate that of last iteration change for remaining ingress to be inserted The insertion cost of one circuit need not be that these nodes compute repeatedly the insertion cost on unchanged route, and the saving time carries High efficiency.In first time iterative process, since all nodes are only there are one insertion position, the sub-optimal of all nodes inserts It is a prodigious integer to enter cost, and the node that insertion Least-cost need to be only found in all nodes is inserted into.
3) selection destroys Reconstruction Strategy combination
Iteration carries out strategy combination selection using wheel disc method each time.Wheel disc method to the degree of different probability selections not Together, it is intended to select Great possibility.The present invention is selected to the corresponding score of each Combination Design, according to score using roulette The method of selecting is selected.Assuming that i-th kind of composite score is πi, have the combination of K kinds, then i-th kind of selected probability of combination is:
4) iteration object is selected
In iterative process each time, select a solution currentSolution as the object for destroying with rebuilding.It is calculating In method implementation procedure, a Solutions set can be safeguarded.I.e. algorithm initialization when, can determine whether set sizes M, deposited in set It puts<=M more excellent solutions randomly choose a solution and are used as current iteration object currentSolution.
Here using random selection without using optimal selection be in order to avoid be absorbed in local search area so as to cause Local optimum.Random selection can increase the randomness of search space, more likely find globally optimal solution.Certainly, here with Machine selection is not blindly to select, several solutions stored in solutions set are current optimal several solutions.
5) it destroys and rebuilds
The destruction Reconstruction Strategy selected based on step 3 destroys reconstruction generation newly to the currentSolution that step 4 selects NewSolution is solved, updates bestEver if newSolution is better than current optimal solution bestEver.
It referring to Fig. 2, is rebuild with selected destruction Reconstruction Strategy to execute to destroy, destroys the part section currently solved first Point removes this part of nodes, then generates new explanation newSolution by repairing this part.
6) acceptance criterion
Algorithm can safeguard that the solutions that a size is M gathers.In an iterative process, if current solutions collection It closes element and is less than M, then directly receive newSolution and update solutions set.Otherwise by newSolution and repeatedly It is compared for object currentSolution, solutions set is updated according to threshold value acceptance criterion.
Threshold value acceptance criterion is to receive the range of poor solution by continuously decreasing and tend to global optimum.Threshold value acceptance criterion Formula is as follows:
costnew< costcurrent+currentThreshold (1)
Present threshold value currentThreshold calculation formula are:
Initial threshold initialThreshold calculation formula are:
InitThreshold=costInitial solution*HRESHOLD_INI (3)
Wherein i is current iteration number, maximum iteration maxIterations, initial threshold coefficient T HRESHOLD_ INI and changes of threshold coefficient alpha are set in algorithm initialization.
7) period scores
In each period that quick self-adapted large neighborhood search algorithm executes, (period that the present embodiment uses is 200 Secondary iteration), not only to count the corresponding score value scoreTotal of each strategy combinationi(score by rules is shown in Table 4) will also count each Execution time exeTime of a strategy within the period in totali。scoreiWhat is indicated is that each strategy combination is held in each period Caused score addition in the row unit interval.
8) weight updates
Weight update is adaptive core procedure, according to institute in each strategy combination executable unit time in each period The score addition score broughtiThe weight of strategy combination is adjusted.The weight of a new round cannot only consider a upper week The iteration situation of phase, historical factor should be also taken into account.
Wherein, ∑ scoreiAfter for current period, caused score in each strategy combination executable unit time Addition scoreiThe sum of.R is the history significance level factor, controls the speed of weight variation.When r is 1, there is no send out for weight It is raw to change, reform into large neighborhood search algorithm;When r is 0, new round weight and each strategy group in current period It is related to close corresponding score value.wijIt is the weight of each strategy combination of upper a cycle, that is, historical factor.
Above formula makes the sum of each strategy combination weight after end cycle be 1, it is therefore an objective to normalize, make weight in (0,1) area In.
Contrast verification is carried out to the present invention below.
Referring to Fig. 3, the figure shows FALNS algorithms Solomon data sets C101, R101, RC101, C102, R102, The change curve of each strategy combination in RC102, it is especially high or even level off to 1 not occur some strategy combination weight, Its strategy combination level off to 0 extreme case.It sees in this way, the strategy combination weight change curve of FALNS is to meet expection 's.
Referring to Fig. 4, what which showed be FALNS algorithms each strategy combination Solomon data sets C101, R101, Weight change curve in RC101, C102, R102, RC102, weight variation of each strategy combination in different data sets Trend illustrates that the quality of strategy combination performance situation depends on specific data set, this is absolutely proved certainly there are prodigious difference The necessity of adaptation.
LNS and FALNS is respectively adopted, 50000 iteration are carried out to 15 data sets of Solomon, it is flat to be repeated 5 times calculating Result is compared, and carrys out the validity of verification algorithm with the comparison of current data set known preferred solution, and result of calculation see the table below 6。
6 LNS and FALNS experimental results of table compare
As can be seen that in addition on RC105 data sets, FALNS solves quality and is slightly worse than LNS, in other 14 data sets On, the solution quality of FALNS has improvement.This illustrates FALNS compared with LNS, has an enormous advantage on solving quality.Separately On the one hand, FALNS can be optimal solution on C101, C102, C103, C104, C105, R105, RC104 this 7 data sets, Even FALNS can improve known preferred solution on R102 data sets.On the solution time of all data sets, FALNS algorithms Half or so even longer time has been saved than LNS algorithm.This synthesis explanation, FALNS can provide for Vehicle Routing Problems While high quality solution, it can substantially shorten than LNS algorithm and solve the time, be more suitable for extensive Vehicle Routing Problems.

Claims (10)

1. a kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems, which is characterized in that packet Include following steps:
Step 1:Parameter training, based on historical data to period scoring institute in quick self-adapted large neighborhood search algorithm The grading parameters being related to are trained;
Step 2:The iterations needed for a cycle are completed in initialization, setting, are generated just using RegretInsert insertions Begin solution;
Step 3:From the combination of multiple destruction Reconstruction Strategies a strategy combination is selected according to wheel disc method;
Step 4:A solution is randomly choosed from current several optimal solution sets Solutions is used as iteration object currentSolution;
Step 5:The destruction Reconstruction Strategy selected in applying step 3 combines the iteration object to being selected in step 4 CurrentSolution carries out destruction reconstruction, generates new explanation newSolution, if newSolution is better than current optimal solution BestEver then updates bestEver with newSolution;
Step 6:Determined whether to receive new explanation newSolution according to threshold value acceptance criterion, updates optimal solution set if receiving Solutions;
Step 7:Judge whether to meet algorithm iteration end condition, step 10 is passed directly to if meeting, otherwise judges current week Whether the phase terminates, and step 3 is gone to if being not over, no to then follow the steps 8;
Step 8:It is scored each strategy combination according to period scoring, primary destroy rebuilds the current optimal solution of generation, Corresponding strategy combination plus s1Point;Primary destroy is rebuild without generating current optimal solution, but is improved than iteration object, corresponding strategy Combination plus s2Point;It is primary to destroy the new explanation rebuild and generated than iteration to aberration, but received by acceptance criterion, corresponding strategy group It closes and adds s3Point;The overall score of each strategy combination in measurement period calculates the score in each strategy combination unit execution time;
Step 9:Each strategy combination history weight and each strategy combination in the new round period are combined according to weight update mechanism Unit executes the score in the time and carries out weight adjustment, then goes to step 3;
Step 10:Export optimal solution bestEver.
2. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that in the step 1, the grading parameters include s1、s2And s3, it is respectively used to quick self-adapted The primary three kinds of different new explanation situation C for destroying and being generated after reconstruction are carried out in large neighborhood search algorithm1、C2And C3It is commented Point, wherein C1It indicates that this time destroys to rebuild and generates current optimal solution, corresponding strategy combination plus s1Point, C2It indicates that this time destroys to rebuild Current optimal solution is not generated, but is improved than iteration object, corresponding strategy combination plus s2Point, C3It indicates that this time destroys and rebuilds life At new explanation than iteration to aberration, but received by acceptance criterion, corresponding strategy combination plus s3Point.
3. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that in the step 2, generate initial solution using RegretInsert insertions, be inserted into be each Node calculates the cost difference of its sub-optimal insertion position and best insertion position, and the maximum node of cost difference is inserted into it most Good insertion position.
4. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that in the step 3, the destruction Reconstruction Strategy be from including RandomRuin, Choose any one kind of them in the destruction algorithm of RadialRuin, WorstRuin, ClusterRuin as destroy the factor, while from including It is chosen any one kind of them in the algorithm for reconstructing of BestInsert, RegretInsert, GreedyInsert as the factor is rebuild, is combined And the destruction Reconstruction Strategy formed.
5. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that in the step 3, select a strategy combination according to wheel disc method, made by following formula Calculative strategy, which is carried out, with the wheel disc method for tending to selection Great possibility combines the probability being chosen to:
Wherein k is the total number of strategy combination, πiIndicate the scoring of i-th kind of strategy combination.
6. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that in the step 4, stored in the optimal solution set Solutions and be not more than working as M Preceding optimal solution.
7. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that in the step 5, pass through the particular content pair for destroying Reconstruction Strategy combination selected in step 3 A part of node currently solved is destroyed, this part of nodes is removed, and then generates new explanation by repairing this part newSolution。
8. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that the step 6 includes following procedure:
In an iterative process, if current solutions set elements are less than M, directly receive newSolution and update Solutions gathers;Otherwise newSolution and iteration object currentSolution are compared, is received according to threshold value Criterion updates solutions set;
The threshold value acceptance criterion is to receive the range of poor solution by continuously decreasing and tend to global optimum, and threshold value receives accurate Then formula is as follows:
costnew< costcurrent+currentThreshold
Wherein costnewFor the expense of original solution, costcurrentFor the expense currently connect;
Present threshold value currentThreshold calculation formula are:
Initial threshold initialThreshold calculation formula are:
InitThreshold=costInitial solution*THRESHOLD_INI
Wherein i is current iteration number, costInitial solutionIt is the expense of initial solution, it is maximum iteration maxIterations, initial Threshold coefficient THRESHOLD_INI and changes of threshold coefficient alpha are set in algorithm initialization.
9. a kind of quick self-adapted extensive neighborhood search towards extensive Vehicle Routing Problems according to claim 1 Method, which is characterized in that when the step 8 executes, within each period, the corresponding score value of each strategy combination need to be counted scoreTotaliWith execution time exeTime of each strategy within the period in totali, and calculate each strategy in each period Combine caused score addition score in executable unit's timei
10. a kind of quick self-adapted extensive neighborhood towards extensive Vehicle Routing Problems according to claim 8 is searched Suo Fangfa, which is characterized in that the weight update mechanism in the step 9 is by for each strategy combination in each period Caused score addition score in executable unit's timeiThe weight of strategy combination is adjusted:
Wherein, ∑ scoreiAfter for current period, caused score addition in each strategy combination executable unit time scoreiThe sum of, r is the history significance level factor, controls the speed of weight variation, r is not less than 0 and is not more than 1, wijOn being The weight of each strategy combination of a cycle, i.e. historical factor.
CN201810355489.2A 2018-04-19 2018-04-19 Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem Active CN108596469B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810355489.2A CN108596469B (en) 2018-04-19 2018-04-19 Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810355489.2A CN108596469B (en) 2018-04-19 2018-04-19 Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem

Publications (2)

Publication Number Publication Date
CN108596469A true CN108596469A (en) 2018-09-28
CN108596469B CN108596469B (en) 2021-11-30

Family

ID=63611323

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810355489.2A Active CN108596469B (en) 2018-04-19 2018-04-19 Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem

Country Status (1)

Country Link
CN (1) CN108596469B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN110097218A (en) * 2019-04-18 2019-08-06 北京邮电大学 Unmanned commodity distribution method and system under changing environment when a kind of
CN111191847A (en) * 2019-12-31 2020-05-22 苏宁云计算有限公司 Distribution path planning method and system considering order polymerization degree
WO2020263177A1 (en) * 2019-06-26 2020-12-30 Swat Mobility Pte. Ltd. Transportation system and method for passenger and route scheduling
CN113935528A (en) * 2021-10-13 2022-01-14 广州市钱大妈农产品有限公司 Intelligent scheduling method and device, computer equipment and storage medium
CN114418773A (en) * 2022-03-30 2022-04-29 支付宝(杭州)信息技术有限公司 Optimization method and device of strategy combination

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103325125A (en) * 2013-07-03 2013-09-25 北京工业大学 Moving target tracking method based on improved multi-example learning algorithm
US20130346057A1 (en) * 2012-06-26 2013-12-26 Eleon Energy, Inc. Methods and systems for power restoration planning
CN107025363A (en) * 2017-05-08 2017-08-08 中国人民解放军国防科学技术大学 A kind of adaptive big neighborhood search method of Agile satellite scheduling
CN107194513A (en) * 2017-05-26 2017-09-22 中南大学 A kind of optimization method for solving full channel logistics distribution

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130346057A1 (en) * 2012-06-26 2013-12-26 Eleon Energy, Inc. Methods and systems for power restoration planning
CN103325125A (en) * 2013-07-03 2013-09-25 北京工业大学 Moving target tracking method based on improved multi-example learning algorithm
CN107025363A (en) * 2017-05-08 2017-08-08 中国人民解放军国防科学技术大学 A kind of adaptive big neighborhood search method of Agile satellite scheduling
CN107194513A (en) * 2017-05-26 2017-09-22 中南大学 A kind of optimization method for solving full channel logistics distribution

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
IMAN DAYARIAN 等: "An adaptive large-neighborhood search heuristic for a multi-period vehicle routing problem", 《TRANSPORTATION RESEARCH PART E: LOGISTICS AND TRANSPORTATION REVIEW 》 *
LUTZ. ROMAN: "Adaptive Large Neighborhood Search", 《HTTPS://OPARU.UNI-ULM.DE/XMLUI/HANDLE/123456789/3264》 *
武勇毅: "混流装配车间车辆路径问题建模和优化", 《中国优秀博硕士学位论文全文数据库(硕士) 经济与管理科学辑》 *
陶杨懿 等: "具有同时服务需求的家庭护理人员调度研究", 《工业工程与管理》 *

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN109635998B (en) * 2018-11-02 2023-04-07 华侨大学 Self-adaptive multi-objective optimization method for solving vehicle path problem with time window
CN110097218A (en) * 2019-04-18 2019-08-06 北京邮电大学 Unmanned commodity distribution method and system under changing environment when a kind of
WO2020263177A1 (en) * 2019-06-26 2020-12-30 Swat Mobility Pte. Ltd. Transportation system and method for passenger and route scheduling
JP2022500796A (en) * 2019-06-26 2022-01-04 スワット、モビリティ、ピーティーイー、リミテッドSWAT Mobility Pte. Ltd. Transportation systems and methods for passenger and route scheduling
JP7057471B2 (en) 2019-06-26 2022-04-19 スワット、モビリティ、ピーティーイー、リミテッド Transportation systems and methods for passenger and route scheduling
CN111191847A (en) * 2019-12-31 2020-05-22 苏宁云计算有限公司 Distribution path planning method and system considering order polymerization degree
CN111191847B (en) * 2019-12-31 2022-10-14 苏宁云计算有限公司 Distribution path planning method and system considering order polymerization degree
CN113935528A (en) * 2021-10-13 2022-01-14 广州市钱大妈农产品有限公司 Intelligent scheduling method and device, computer equipment and storage medium
CN114418773A (en) * 2022-03-30 2022-04-29 支付宝(杭州)信息技术有限公司 Optimization method and device of strategy combination

Also Published As

Publication number Publication date
CN108596469B (en) 2021-11-30

Similar Documents

Publication Publication Date Title
CN108596469A (en) A kind of quick self-adapted extensive neighborhood search method towards extensive Vehicle Routing Problems
Fischetti et al. The generalized traveling salesman and orienteering problems
CN111798067B (en) Automatic driving automobile distribution path planning method based on self-adaptive large neighborhood search algorithm
CN107194513A (en) A kind of optimization method for solving full channel logistics distribution
CN109391645A (en) Block chain light-weight technologg method, block chain node and storage medium
CN105974799A (en) Fuzzy control system optimization method based on differential evolution-local unimodal sampling algorithm
CN104200263A (en) Power distribution network route planning method based on tabu differential evolution and GIS (Geographic Information System)
CN104036324A (en) Optimal design method of communication network capacity based on genetic algorithm
CN101677286A (en) Optimization method of carrier network
CN105356465B (en) Transmission Expansion Planning in Electric platform and application that a kind of economy is coordinated with security
Malik et al. A branch and bound algorithm for the capacitated minimum spanning tree problem
CN110225096B (en) Server load balancing method based on genetic algorithm
CN111047439A (en) Transaction processing method based on block chain
CN110276689B (en) Intelligent contract implementation method based on dynamic decision
CN110427690A (en) A kind of method and device generating ATO rate curve based on global particle swarm algorithm
CN109617702A (en) Method, block chain node and the device with store function of information signature
CN114567554B (en) Block chain construction method based on node reputation and partition consensus
CN104168570B (en) Authenticity dynamic bidirectional spectrum auction method based on interference information
CN104881549A (en) Power consumption optimizing method for RM (Reed-Muller) logical circuit comprising irrelevant items
CN109635998A (en) A kind of adaptive Multipurpose Optimal Method solving vehicle routing problem with time windows
CN106919504A (en) A kind of test data evolution generation method based on IMPE GA algorithms
CN104392317A (en) Project scheduling method based on genetic culture gene algorithm
CN106682729A (en) BP neural network MapReduce training method based on local convergence weight matrix evolution
CN105589896B (en) Data digging method and device
CN104283214B (en) Network reconstruction method for power distribution network

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
GR01 Patent grant
GR01 Patent grant