CN108596469B - Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem - Google Patents

Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem Download PDF

Info

Publication number
CN108596469B
CN108596469B CN201810355489.2A CN201810355489A CN108596469B CN 108596469 B CN108596469 B CN 108596469B CN 201810355489 A CN201810355489 A CN 201810355489A CN 108596469 B CN108596469 B CN 108596469B
Authority
CN
China
Prior art keywords
strategy
strategy combination
solution
reconstruction
current
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.)
Active
Application number
CN201810355489.2A
Other languages
Chinese (zh)
Other versions
CN108596469A (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

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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0637—Strategic 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
    • G06Q10/06375—Prediction 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)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Navigation (AREA)

Abstract

The invention discloses a rapid self-adaptive large-scale neighborhood searching method facing to a large-scale vehicle path problem, which is characterized in that a period scoring mechanism and a strategy combination weight adjusting mechanism are added, N times of iteration is taken as a period, weight adjustment is carried out according to the performance quality condition in the execution time of each strategy combination unit in the period, the probability of selecting a strategy combination is increased by improving the weight of the strategy combination with good performance, and the probability of selecting the strategy combination is reduced by reducing the weight of the strategy combination with poor performance. Along with the deepening of iteration, the strategy combination which expresses better in unit execution time has higher probability to be selected, and the quick self-adaption is realized.

Description

Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem
Technical Field
The invention relates to a rapid self-adaptive large-scale neighborhood searching method for a large-scale vehicle path problem.
Background
The vehicle path problem is generally defined as: the vehicles start from one or more central points to provide services for a plurality of customer points with different cargo demands, and a proper driving route is planned under the conditions of meeting the demands of the customer points and certain constraint conditions, so that the aim of lowest transportation cost is fulfilled.
The vehicle path problem mainly comprises two main solving methods, namely an accurate algorithm and a heuristic algorithm. For smaller-scale vehicle path problems, accurate algorithms such as branch definition algorithms and dynamic planning algorithms can be successfully solved, but the solving time of the vehicle path problems is exponentially increased along with the increase of the problem scale, so that the accurate algorithms are difficult to solve the large-scale vehicle path problems. Therefore, heuristic algorithms have been studied, of which the reduction method, the scanning method, the two-stage method, and the like are most representative. In recent years, heuristic algorithms are rapidly developed, and modern optimization algorithms such as genetic algorithms, simulated annealing algorithms, tabu search algorithms, large-scale neighborhood search algorithms and the like are also applied to solving the vehicle path problem.
The large-scale neighborhood search algorithm is an extension of a destructive reconstruction algorithm, and the main idea is to construct a plurality of destructive factors (table 3) and reconstruction factors (table 4) according to a destructive reconstruction principle. The destruction factors and the reconstruction factors are combined pairwise to form destruction reconstruction strategy combinations (table 1), certain weight is set for each strategy combination when an algorithm is initialized, one strategy combination is selected according to a certain strategy in each step of iteration process to destroy and reconstruct an iteration object to generate a new solution, whether the new solution is accepted or not is judged according to an acceptance criterion, the current optimal solution is stored, and the optimal solution is output after iteration circulation is carried out until an iteration termination condition is met. In the iteration process, each strategy combination is selected with a certain probability, and compared with a destructive reconstruction algorithm, the method has a larger search space, has less possibility of falling into local optimum, increases the possibility of obtaining a global optimum solution, and is called a large-scale field search algorithm. In each iteration process, the probability of each strategy combination being selected is closely related to the weight of the strategy combination, but in the large-scale neighborhood searching algorithm, the weight of each strategy combination is determined in the process of initializing the algorithm and is not changed in the process of the algorithm. Therefore, the flexibility of the algorithm is low, the solving time is long, and the solving effect is not ideal.
Disclosure of Invention
The invention aims to provide a Fast Adaptive Large-scale Neighborhood searching Algorithm (FaLNS) on the basis of a Large-scale Neighborhood searching Algorithm (LNS), and solves the problem that the original traditional Algorithm cannot provide a high-quality feasible solution in a short time due to the increase of distribution scale in the actual logistics distribution process of an enterprise.
In order to achieve the technical purpose, the technical scheme of the invention is that,
a quick self-adaptive large-scale neighborhood searching method for a large-scale vehicle path problem comprises the following steps:
step 1: parameter training, namely training scoring parameters related to a periodic scoring mechanism in a rapid self-adaptive large-scale neighborhood searching algorithm based on historical data;
step 2: initializing, setting iteration times required for completing one period, and generating an initial solution by using a RegretInsert insertion method;
and step 3: selecting a strategy combination from a plurality of destroyed reconstruction strategy combinations according to a wheel disc method;
and 4, step 4: randomly selecting one solution from a plurality of current optimal solution sets as an iteration object currentSolution;
and 5: performing destructive reconstruction on the iteration object currentSolution selected in the step 4 by applying the destructive reconstruction strategy combination selected in the step 3 to generate a new solution newSolution, and updating the bestEver by the newSolution if the newSolution is better than the current optimal solution bestEver;
step 6: judging whether to accept the new solution according to a threshold acceptance criterion, and updating the optimal solution set Solutions if the new solution is accepted;
and 7: judging whether an algorithm iteration termination condition is met, and if so, directly turning to the step 10; otherwise, judging whether the current period is finished, if not, turning to the step 3, otherwise, executing the step 8, and if not, turning to the step 3. The iteration termination condition may be set to satisfy a preset maximum iteration number, satisfy an improvement ratio of the solution, and the like. And if the iteration termination condition is not met, continuing to iterate. Before continuing to iterate, whether the current cycle is finished or not is judged, namely whether the preset iteration times in one cycle are finished or not is judged, so that the strategy combination is scored for the current cycle.
And 8: and scoring each strategy combination according to a periodic scoring mechanism. Reconstructing once destruction to generate current optimal solution, and adding s to corresponding strategy combination1Dividing; the one-time destructive reconstruction does not generate the current optimal solution, but is improved compared with the iterative object, and the corresponding strategy combination is added with s2Dividing; the new solution ratio iteration generated by the one-time destructive reconstruction is opposite to the aberration, but is accepted by the acceptance criterion, and the corresponding strategy combination is added with s3And (4) dividing. Counting the total scores of all the strategy combinations in a period, and calculating the score in the execution time of each strategy combination unit;
and step 9: performing weight adjustment according to a weight updating mechanism by combining historical weights of all strategy combinations and scores in execution time of all strategy combination units in a new cycle, and then turning to the step 3;
step 10: and outputting the optimal solution bestEver.
In the step 1, the scoring parameters include s1、s2And s3Respectively used for three different new solution conditions C generated after once destructive reconstruction in the fast self-adaptive large-scale neighborhood search algorithm1、C2And C3Scoring, wherein C1Representing the destruction reconstruction to generate the current optimal solution, and adding s to the corresponding strategy combination1Is divided into C2The method shows that the destructive reconstruction does not generate the current optimal solution, but is improved compared with the iterative object, and the corresponding strategy combination adds s2Is divided into C3The new solution ratio iteration generated by the destructive reconstruction is expressed to aberration, but is accepted by the acceptance criterion, and the corresponding strategy combination is added with s3And (4) dividing. If the generated new solution is not accepted by the acceptance criteria, no scoring is performed.
In the step 2, an initial solution is generated by using a regret insert method, wherein a cost difference value between a next best insert position and a best insert position of each node to be inserted is calculated, and the node with the largest cost difference value is inserted into the best insert position.
In the step 3, the destructive reconstruction strategy is a destructive reconstruction strategy formed by combining any one of destructive algorithms including randomrun, RadialRuin, worstrun and clusterrun as a destructive factor and any one of reconstruction algorithms including bestsert, RegretInsert and greeny insert as a reconstruction factor.
In the step 3, a strategy combination is selected according to a wheel disc method, and the probability of the selected strategy combination is calculated by using the wheel disc method which tends to select a high probability event through the following expression:
Figure BDA0001634591330000041
where k is the total number of strategy combinations, πiRepresents the score of the ith combination of strategies.
In the step 4, no more than M current optimal Solutions are stored in the optimal solution set Solutions. Wherein M is a preset parameter, that is, the top M solutions with the highest scores are directly taken as the current optimal solution set.
In the step 5, a part of nodes of the current solution is destroyed through specific contents of the destroyed reconstruction strategy combination selected in the step 3, the part of nodes is removed, and then a new solution newSolution is generated by repairing the part of nodes.
The fast self-adaptive large-scale neighborhood searching method facing to the large-scale vehicle path problem comprises the following steps of 6:
in the iteration process, if the current solutions set elements are smaller than M, directly receiving newsolutions and updating the solutions set; otherwise, comparing the newSolution with the iterative object currentSolution, and updating the solutions set according to the threshold acceptance criterion;
the threshold acceptance criterion tends to be global optimal by gradually reducing the range of accepting poor solutions, and the formula of the threshold acceptance criterion is as follows:
costnew＜costcurrent+currentThreshold
wherein costnewCost for the original solutioncurrentThe overhead for the current connection;
the current threshold currentThreshold is calculated as:
Figure BDA0001634591330000051
the initial threshold initialThreshold is calculated by the formula:
initThreshold＝costinitial solution*THRESHOLD_INI
Where i is the current iteration number, costInitial solutionIs the cost of the initial solution, and the maximum number of iterations maxtiterations, the initial THRESHOLD coefficient THRESHOLD _ INI, and the THRESHOLD variation coefficient alpha are set at the time of algorithm initialization.
In the method for fast self-adaptive large-scale neighborhood searching for the large-scale vehicle path problem, when the step 8 is executed, the score scoreTotal corresponding to each strategy combination needs to be counted in each periodiAnd the total execution time exeTime of each policy in the periodiAnd calculating the score addition score brought by each strategy combination execution unit time in each periodi：
Figure BDA0001634591330000061
In the method for fast adaptive large-scale neighborhood search for the large-scale vehicle path problem, the weight updating mechanism in the step 9 is score addition score brought by each strategy combination in each period in unit timeiAnd adjusting the weight of the strategy combination:
Figure BDA0001634591330000062
wherein, sigma scoreiAfter the current period is finished, the score added in unit time is executed by each strategy combinationiThe sum, r, is a historical importance factor, controlling the speed of weight change. r is not less than 0 and not more than 1. When r is 1, the weight is not changed, and the large-scale neighborhood searching algorithm is realized; when r is 0, the new round of weight is only related to the scores corresponding to the strategy combinations in the current period,experiments show that the value of r is suitable (0.9, 0.95). w is aijIs the weight of each strategy combination in the last period, namely the historical factor.
Compared with a large-scale neighborhood search algorithm, the method has the advantages that the method can provide a high-quality solution for the vehicle path problem, can greatly shorten the solving time compared with the large-scale neighborhood search algorithm, and is more suitable for the large-scale vehicle path problem.
The invention will be further explained with reference to the drawings.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of a destructive reconstruction principle;
FIG. 3 is a schematic diagram of the variation curve of each destruction and reconstruction strategy combination of a single data set according to the present invention;
fig. 4 is a schematic diagram of variation curves of a single destructive reconstruction strategy combination in different data sets according to the present invention.
Detailed Description
The rapid self-adaptive large-scale neighborhood searching algorithm provided by the invention is an extension of the large-scale neighborhood searching algorithm. In the large-scale neighborhood searching algorithm, the weight of each strategy combination cannot be adjusted after the algorithm is initialized, and a period scoring mechanism and a strategy combination weight adjusting mechanism are added into the self-adaptive large-scale neighborhood searching algorithm. Taking N iterations as a period, where N takes a value of 200 in this embodiment, the weight adjustment is performed according to the performance quality in the execution time of each policy combination unit in the period, the probability of selecting the policy combination is increased by increasing the weight of the policy combination that performs well, and the probability of selecting the policy combination is decreased by decreasing the weight of the policy combination that performs poorly. Along with the deepening of iteration, the strategy combination which expresses better in unit execution time has higher probability to be selected, and the quick self-adaption is realized.
Referring to fig. 1, the present invention comprises the steps of:
step 1: parameter training, namely training scoring parameters related to a periodic scoring mechanism in the fast self-adaptive large-scale neighborhood searching algorithm by using relevant historical data based on the large-scale neighborhood searching algorithm;
step 2: initializing, and generating an initial solution by utilizing a RegretInsert insertion method;
and step 3: selecting a strategy combination from a plurality of destroyed reconstruction strategy combinations according to a wheel disc method; specific combinations of destruction reconstruction strategies can be seen in table 1 below:
TABLE 1 policy combination Table
Figure BDA0001634591330000071
Examples of partial policy combinations are given in the above table, and in practice, the reconstruction policy is destroyed by using any one of the destroying algorithms including randomrun, radialrain, worstrun, clusterrun as a destroying factor, and by using any one of the reconstructing algorithms including bestlert, RegretInsert, greeny insert as a reconstructing factor, i.e., more different policy combinations can be formed, which are not all expressed in the table.
And 4, step 4: randomly selecting one solution from a plurality of current optimal solution sets as an iteration object currentSolution;
and 5: referring to fig. 2, the destruction reconstruction of the iterative object currentSolution selected in step 4 is performed by applying the destruction reconstruction policy combination selected in step 3, so as to generate a new solution newSolution, and whether to update the current optimal solution bestEver is judged; the common failure phase algorithm can be seen in table 2 below:
TABLE 2 Scoring rules Table
Figure BDA0001634591330000081
A commonly used reconstruction phase algorithm can be seen in table 3 below:
TABLE 3 common reconstruction phase Algorithm
Figure BDA0001634591330000082
Step 6: judging whether to accept the new solution according to a threshold acceptance criterion, and updating the optimal solution set Solutions if the new solution is accepted;
and 7: and judging whether an algorithm iteration termination condition is met, if so, directly turning to the step 10, otherwise, judging whether the current period is finished, if not, turning to the step 3, and otherwise, executing the step 8. The algorithm iteration termination condition in this embodiment is set to satisfy a preset iteration number, and the specific number is 50000.
And 8: scoring each strategy combination according to a periodic scoring mechanism, once destroying and reconstructing to generate a current optimal solution, and adding s to the corresponding strategy combination1Dividing; the one-time destructive reconstruction does not generate the current optimal solution, but is improved compared with the iterative object, and the corresponding strategy combination is added with s2Dividing; the new solution ratio iteration generated by the one-time destructive reconstruction is opposite to the aberration, but is accepted by the acceptance criterion, and the corresponding strategy combination is added with s3And (4) dividing. Counting the total scores of all the strategy combinations in a period, and calculating the score in the execution time of each strategy combination unit; see table 4 below for fine scores:
TABLE 4 Scoring rules Table
Figure BDA0001634591330000091
And step 9: performing weight adjustment according to a weight updating mechanism by combining historical weights of all strategy combinations and scores in execution time of all strategy combination units in a new cycle, and then turning to the step 3;
step 10: and outputting the optimal solution bestEver.
1) Training parameters
The invention introduces a periodic scoring mechanism to score each iteration in the period, so s needs to be determined in advance1、s2、s3And taking values of the parameters. The invention utilizes historical data to s based on large-scale neighborhood search algorithm1、s2、s3And (5) carrying out training by parameter value taking. Table 5 below shows the results of experiments on 15 Solomon datasets, where C is counted in 10000 iterations for each dataset1、C2、C3The number of 3 conditions appeared, and the difficulty of the 3 conditions appeared was related to s1、s2、s3Taking values:
TABLE 5 comparison of the number of occurrences of the three conditions
Figure BDA0001634591330000101
TABLE 5 by C1Based on the number of occurrences, C was calculated separately for each data set2、C3Number of occurrences of two conditions and C1Then, after removing the two highest and lowest ratios, respectively, for a total of 4 ratios, an average is calculated based on the remaining 11 ratios to obtain an approximate ratio. It can be seen that C3The number of occurrences is probably C125 times of that of, C2The number of occurrences is probably C15 times of the total weight of the powder. The difficulty of the 3 conditions should be inversely proportional to the corresponding score, so if s3Is 1, then s2And s35 and 25 respectively. The data set of the training parameters is the same as the data set of the subsequent experiment, and the coupling degree is higher to a certain degree. However, in the actual case of an enterprise, it is realistic to train appropriate parameters using historical data as a basis for routine data processing.
2) Initialization
The present invention uses the RegretInsert algorithm to generate an initial solution. The generation process of the initial solution is an iterative process, and each iteration finds one node to carry out insertion operation. The regret insert algorithm calculates the cost difference value between the next best insertion position and the best insertion position of each node to be inserted in each iteration process, inserts the node with the largest cost difference value into the best insertion position (if only the best insertion position is available, and the cost of the next best insertion position is not set to be a large integer), and in the next iteration process, only the insertion cost of the line which is changed by the last iteration needs to be calculated again for the rest nodes to be inserted, and the insertion cost on the unchanged line does not need to be calculated repeatedly for the nodes, so that the time is saved, and the efficiency is improved. In the first iteration process, all nodes have only one insertion position, so that the sub-optimal insertion cost of all nodes is a large integer, and only the node with the minimum insertion cost needs to be found from all nodes for insertion.
Figure BDA0001634591330000111
3) Selective destruction reconstruction strategy combination
And carrying out strategy combination selection by using a wheel disc method in each iteration. The degree of selection of different probabilities by the roulette method tends to select high probability events. The invention designs corresponding scores for each combination, and selects by adopting a roulette selection method according to the scores. Suppose that the ith combination score is piiIf there are K combinations, the probability of the ith combination being selected is:
Figure BDA0001634591330000121
4) selecting an iterative object
In each iteration process, one solution currentSolution is selected as the object of destruction and reconstruction. During algorithm execution, a set of Solutions is maintained. That is, when the algorithm is initialized, the size M of the set is determined, M better solutions are stored in the set, and one solution is randomly selected as the currentSolution of the current iteration object.
The random selection is used here without the optimal selection in order to avoid trapping in local search areas resulting in local optima. Random selection may increase the randomness of the search space, making it more likely that a globally optimal solution will be found. Of course, the random selection here is not blind selection, and several solutions stored in the solutions set are currently the best ones.
5) Destructive reconstruction
And (4) generating a new solution newSolution for the currentSolution destructive reconstruction selected in the step (4) based on the destructive reconstruction strategy selected in the step (3), and updating the bestEver if the newSolution is superior to the best solution bestEver at present.
Referring to fig. 2, the destructive reconstruction is performed with the selected destructive reconstruction strategy by first destroying a portion of the nodes of the current solution, removing the portion of nodes, and then generating a new solution newSolution by repairing the portion.
6) Acceptance criteria
The algorithm maintains a set of conditions of size M. In the iterative process, if the current solutions set elements are smaller than M, the newsolutions are directly accepted and the solutions set is updated. Otherwise, comparing the newSolution with the iterative object currentSolution, and updating the solutions set according to the threshold acceptance criterion.
The threshold acceptance criterion is to tend to be globally optimal by gradually reducing the range of accepting poor solutions. The threshold acceptance criterion formula is as follows:
costnew＜costcurrent+currentThreshold (1)
the current threshold currentThreshold is calculated as:
Figure BDA0001634591330000131
the initial threshold initialThreshold is calculated by the formula:
initThreshold＝costinitial solution*HRESHOLD_INI (3)
Wherein i is the current iteration number, and the maximum iteration number maxIterations, the initial THRESHOLD coefficient THRESHOLD _ INI and the THRESHOLD variation coefficient alpha are set during algorithm initialization.
7) Periodic scoring
In each period (the period adopted in this embodiment is 200 iterations) of the fast adaptive large-scale neighborhood search algorithm, not only the score scoreTotal corresponding to each policy combination needs to be countedi(see Table 4 for details of the scoring), and the total execution time exeTime of each strategy in the period is countedi。scoreiThe point addition brought by each strategy combination in each period in unit time is shown.
Figure BDA0001634591330000132
8) Weight update
The weight updating is a self-adaptive core step, and score addition score brought by each strategy combination in each period in unit time is executediThe weights of the policy combinations are adjusted. The new round of weights cannot only take into account the iteration case of the previous cycle, and historical factors should also be taken into account.
Figure BDA0001634591330000133
Wherein, sigma scoreiAfter the current period is finished, the score added in unit time is executed by each strategy combinationiAnd (4) summing. r is a historical importance factor that controls the speed at which the weight changes. When r is 1, the weight is not changed, and the large-scale neighborhood searching algorithm is realized; when r is 0, the new round of weight is only related to the scores corresponding to the strategy combinations in the current period. w is aijIs the weight of each policy combination in the last cycle, i.e. the historical factor.
Figure BDA0001634591330000141
The above formula makes the sum of the weights of each strategy combination after the period is over 1, and aims to perform normalization so that the weights are in the (0,1) interval.
The invention is now verified by comparison.
Referring to fig. 3, the graph shows the variation curves of the FALNS algorithm in the Solomon data sets C101, R101, RC101, C102, R102, RC102, and there is no extreme case that a certain policy combination has a particularly high weight or even approaches to 1, and other policy combinations approach to 0. In this regard, the policy combination weight variation curve of FALNS is expected.
Referring to fig. 4, the graph shows weight variation curves of various policy combinations of the FALNS algorithm in the Solomon data sets C101, R101, RC101, C102, R102, and RC102, and there is a great difference in weight variation trend of the various policy combinations in different data sets, which indicates how well the performance of the policy combinations depends on specific data sets, which fully illustrates the necessity of adaptation.
Respectively adopting LNS and FALNS to carry out 50000 times of iteration on 15 data sets of Solomon, repeating 5 times of calculation average results for comparison, and comparing with the known optimal solution of the current data set to verify the effectiveness of the algorithm, wherein the calculation results are shown in the following table 6.
TABLE 6 comparison of LNS and FALNS Experimental results
Figure BDA0001634591330000142
Figure BDA0001634591330000151
It can be seen that the FALNS solution quality was slightly worse than LNS on the RC105 data set, and improved on the other 14 data sets. This shows that the FALNS has a great advantage in solution quality compared to the LNS. On the other hand, the FALNS can achieve the optimal solution on 7 data sets of C101, C102, C103, C104, C105, R105 and RC104, and the FALNS can improve the known optimal solution even on the R102 data set. The FALNS algorithm saves about half or even more time than the LNS algorithm over the solution time of all data sets. This demonstrates that the FALNS can provide a high quality solution for the vehicle routing problem while significantly reducing the solution time compared to the LNS algorithm, and is more suitable for large-scale vehicle routing problems.

Claims (7)

1. A quick self-adaptive large-scale neighborhood searching method for a large-scale vehicle path problem is applied to the actual logistics distribution process of an enterprise, and is characterized by comprising the following steps of:
step 1: parameter training, namely training scoring parameters related to a periodic scoring mechanism in a rapid self-adaptive large-scale neighborhood searching algorithm based on historical data;
step 2: initializing, setting iteration times required for completing one period, and generating an initial solution by using a RegretInsert insertion method;
and step 3: selecting a strategy combination from a plurality of destroyed reconstruction strategy combinations according to a wheel disc method;
and 4, step 4: randomly selecting one solution from a plurality of current optimal solution sets as an iteration object currentSolution;
and 5: performing destructive reconstruction on the iteration object currentSolution selected in the step 4 by applying the destructive reconstruction strategy combination selected in the step 3 to generate a new solution newSolution, and updating the bestEver by the newSolution if the newSolution is better than the current optimal solution bestEver;
step 6: judging whether to accept the new solution according to a threshold acceptance criterion, and updating the optimal solution set Solutions if the new solution is accepted;
and 7: judging whether an algorithm iteration termination condition is met, if so, directly turning to the step 10, otherwise, judging whether the current period is finished, if not, turning to the step 3, otherwise, executing the step 8;
and 8: scoring each strategy combination according to a periodic scoring mechanism, once destroying and reconstructing to generate a current optimal solution, and adding s to the corresponding strategy combination1Dividing; the one-time destructive reconstruction does not generate the current optimal solution, but is improved compared with the iterative object, and the corresponding strategy combination is added with s2Dividing; the new solution ratio iteration generated by the one-time destructive reconstruction is opposite to the aberration, but is accepted by the acceptance criterion, and the corresponding strategy combination is added with s3Dividing; counting the total scores of all the strategy combinations in a period, and calculating the score in the execution time of each strategy combination unit;
and step 9: performing weight adjustment according to a weight updating mechanism by combining historical weights of all strategy combinations and scores in execution time of all strategy combination units in a new cycle, and then turning to the step 3;
step 10: outputting an optimal solution bestEver;
in step 3, selecting a strategy combination according to the roulette method, calculating the probability of the selected strategy combination by using the roulette method which tends to select a high probability event according to the following expression:
Figure FDA0003289327170000021
where k is the total number of strategy combinations, πiA score representing the ith combination of strategies;
when the step 8 is executed, in each period, the score scoreTotal corresponding to each strategy combination needs to be countediAnd the total execution time exeTime of each policy in the periodiAnd calculating the score addition score brought by each strategy combination execution unit time in each periodi：
Figure FDA0003289327170000022
The weight updating mechanism in step 9 is a score added score brought by each policy combination execution unit time in each periodiAnd adjusting the weight of the strategy combination:
Figure FDA0003289327170000023
wherein, sigma scoreiAfter the current period is finished, the score added in unit time is executed by each strategy combinationiThe sum of the weights is a history importance factor controlling the weight change speed, r is not less than 0 and not more than 1, wijIs the weight of each strategy combination in the last period, namely the historical factor.
2. The method for fast adaptive large-scale neighborhood search for large-scale vehicle routing problem according to claim 1, wherein said evaluating in step 1The sub-parameters include s1、s2And s3Respectively used for three different new solution conditions C generated after once destructive reconstruction in the fast self-adaptive large-scale neighborhood search algorithm1、C2And C3Scoring, wherein C1Representing the destruction reconstruction to generate the current optimal solution, and adding s to the corresponding strategy combination1Is divided into C2The method shows that the destructive reconstruction does not generate the current optimal solution, but is improved compared with the iterative object, and the corresponding strategy combination adds s2Is divided into C3The new solution ratio iteration generated by the destructive reconstruction is expressed to aberration, but is accepted by the acceptance criterion, and the corresponding strategy combination is added with s3And (4) dividing.
3. The fast adaptive large-scale neighborhood searching method facing the large-scale vehicle path problem according to claim 1, wherein in the step 2, the RegretInsert interpolation method is used to generate the initial solution, the cost difference between the next best insertion position and the best insertion position of each node to be inserted is calculated, and the node with the largest cost difference is inserted into the best insertion position.
4. The method as claimed in claim 1, wherein in step 3, the reconstruction destroying strategy is a combined reconstruction destroying strategy formed by selecting one of destroying algorithms including randomrun, radialrain, worstrun and clusterrun as a destroying factor, and selecting one of reconstructing algorithms including bestsert, RegretInsert and greeny insert as a reconstructing factor.
5. The method as claimed in claim 1, wherein in step 4, the optimal solution set Solutions store no more than M current optimal Solutions.
6. The method for fast adaptive large-scale neighborhood search facing large-scale vehicle path problem according to claim 1, wherein in step 5, a part of nodes of the current solution is destroyed by the specific content of the destroyed reconstruction strategy combination selected in step 3, and the part of nodes is removed, and then a new solution is generated by repairing the part of nodes.
7. The method for fast adaptive large-scale neighborhood searching for large-scale vehicle path problem according to claim 5, wherein the step 6 comprises the following steps:
in the iteration process, if the current solutions set elements are smaller than M, directly receiving newsolutions and updating the solutions set; otherwise, comparing the newSolution with the iterative object currentSolution, and updating the solutions set according to the threshold acceptance criterion;
the threshold acceptance criterion tends to be global optimal by gradually reducing the range of accepting poor solutions, and the formula of the threshold acceptance criterion is as follows:
costnew＜costcurrent+currentThreshold
wherein costnewCost for the original solutioncurrentThe overhead for the current connection;
the current threshold currentThreshold is calculated as:
Figure FDA0003289327170000041
the initial threshold initialThreshold is calculated by the formula:
initThreshold＝costinitial solution*THRESHOLD_INI
Where i is the current iteration number, costInitial solutionIs the cost of the initial solution, and the maximum number of iterations maxtiterations, the initial THRESHOLD coefficient THRESHOLD _ INI, and the THRESHOLD variation coefficient alpha are set at the time of algorithm initialization.
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 CN108596469A (en) 2018-09-28
CN108596469B true 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)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN110097218B (en) * 2019-04-18 2021-04-13 北京邮电大学 Unmanned commodity distribution method and system in time-varying environment
SG10201905938QA (en) * 2019-06-26 2021-01-28 Swat Mobility Pte Ltd Transportation System and Method for passenger and route scheduling
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
CN113935528B (en) * 2021-10-13 2023-07-21 广州市钱大妈信息科技有限公司 Intelligent scheduling method, intelligent scheduling 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 (3)

* 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
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

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9300134B2 (en) * 2012-06-26 2016-03-29 Eleon Energy, Inc. Methods and systems for power restoration planning

Patent Citations (3)

* 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
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
Adaptive Large Neighborhood Search;Lutz. Roman;《https://oparu.uni-ulm.de/xmlui/handle/123456789/3264》;20150528;1-107 *
An adaptive large-neighborhood search heuristic for a multi-period vehicle routing problem;Iman Dayarian 等;《Transportation Research Part E: Logistics and Transportation Review 》;20161130;第95卷;95–123 *
具有同时服务需求的家庭护理人员调度研究;陶杨懿 等;《工业工程与管理》;20170610;第22卷(第3期);120-127、143 *
混流装配车间车辆路径问题建模和优化;武勇毅;《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》;20150415(第4期);J150-203 *

Also Published As

Publication number Publication date
CN108596469A (en) 2018-09-28

Similar Documents

Publication Publication Date Title
CN108596469B (en) Rapid self-adaptive large-scale neighborhood searching method for large-scale vehicle path problem
CN108053119B (en) Improved particle swarm optimization method for solving scheduling problem of zero-waiting line shop
CN112380008B (en) Multi-user fine-grained task unloading scheduling method for mobile edge computing application
Cordone et al. A heuristic for the vehicle routing problem with time windows
JP6724576B2 (en) Multi-target optimization method and apparatus
CN111798067B (en) Automatic driving automobile distribution path planning method based on self-adaptive large neighborhood search algorithm
CN113361201B (en) Crowd-sourced acquired tag data cleaning method based on noise tag learning
CN113296496B (en) Gravity self-adaptive step length bidirectional RRT path planning method based on multiple sampling points
CN115935442A (en) Block chain performance optimization method based on multi-agent deep reinforcement learning
CN111277277A (en) Method and device for reducing decoding delay of polar code continuous cancellation table decoding algorithm
CN112800384A (en) GTSP solving algorithm based on self-adaptive large-field search
CN110659735A (en) Method, device and equipment for dynamically adjusting neural network channel
CN104881549A (en) Power consumption optimizing method for RM (Reed-Muller) logical circuit comprising irrelevant items
CN108415774B (en) Software and hardware partitioning method based on improved firework algorithm
US20080147579A1 (en) Discriminative training using boosted lasso
CN110942184B (en) Self-adaptive addressing route-finding planning method, device, equipment and storage medium
CN112183762A (en) Reinforced learning method based on mixed behavior space
CN116432838A (en) Wind power prediction method, system, equipment and medium based on VMD-HS-TimesNet combination model
CN110516806A (en) The rarefaction method and apparatus of neural network parameter matrix
CN113689001B (en) Virtual self-playing method and device based on counter-facts regretation minimization
CN113112031A (en) Unmanned aerial vehicle task allocation method based on simulated annealing algorithm
CN111723834A (en) Voice deep learning training method and device
CN108181938B (en) Water level adjusting method for reversely dissolving waste water of hydropower station
JP3847317B2 (en) Optimal solution search device
CN114415615B (en) Mixed flow assembly line balanced distribution method and device under uncertain demand

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