TWI652629B - Mixed genetic algorithm - Google Patents

Abstract

The present invention provides a hybrid genetic algorithm method, comprising the steps of: receiving a tracking target to randomly generate a complex solution path; converting each solution path into a first adaptation target; comparing the first adaptation target with the pursuit target, generating a non- In the solution, the first adaptation target is subjected to a genetic operation to generate a plurality of second adaptation targets; the second adaptation target is subjected to a single point search operation and a comparison operation to generate a plurality of first evaluation paths and second evaluation paths. Converting each first evaluation path and each second evaluation path into a mixed adaptation target; when the mixed adaptation target meets the pursuit target, the mixed adaptation target is regarded as an optimal solution, thereby achieving fast convergence and reducing the calculation time effect .

Description

Hybrid genetic algorithm

The invention relates to a calculation method, in particular to a hybrid genetic operation method.

Genetic algorithm (GA) is a search algorithm used in computational mathematics to solve optimization. The genetic algorithm retains better offspring through the process of initial parent evolution and iteration.

Gene algorithm flow: randomly generate n chromosomes at the beginning; use adaptive function to calculate the fitness values of all chromosomes; use adaptive function to calculate the fitness values of all chromosomes; use adaptive function to calculate the fitness values of all chromosomes; repeat the above "assessment adaptive function" Steps such as "selection, copying", "mating", and "mutation" are performed 2 to 4 times. The operation of performing the above steps 2 to 4 times is called 1 iteration until convergence, and the condition of convergence is that The number of generations reaches a certain number of times or all chromosomes are very similar.

Although the genetic algorithm has global search ability, it can search out the whole solution in the solution space without falling into the problem of rapid decline of local solutions. However, the local search ability of the genetic algorithm is poor, which makes the simple genetic algorithm more time-consuming. In the late stage of evolution, the search efficiency is low. In practical applications, the genetic algorithm is prone to the problem of premature convergence. It is not necessarily that the best solution is found and converges, causing miscalculations.

In order to solve the above problems, the present invention provides a hybrid genetic algorithm method, which combines a search target with a single point search operation and a comparison operation through a genetic operation to achieve a fast convergence and optimization result, thereby reducing the calculation time.

An embodiment of the present invention provides a hybrid genetic algorithm method, including the steps of: receiving a tracking target, randomly generating a complex solution path for the pursuit target; converting each solution path into a corresponding one of the first adaptation targets; When the first adaptation target is compared with the pursuit target, and the infeasible solution is generated, the first adaptation target is subjected to a genetic operation to generate a plurality of second adaptation targets; and the second adaptation target is performed as a single Point search operation to generate a plurality of first evaluation paths; performing a comparison operation on the second adaptation target to generate a plurality of second evaluation paths; converting each first evaluation path and each second evaluation path into a mixed adaptive target And comparing the mixed adaptation target with the pursuit target, and when the mixed adaptation target meets the pursuit target, the mixed adaptation target is regarded as the optimal solution.

In one of the embodiments, the single point search operation compares each of the second adaptive targets with the pursuit target one by one to generate respective first evaluation paths that meet the pursuit target.

In one of the embodiments, the single point search operation is a simulated annealing method.

In one of the embodiments, the comparing operation stores the second adaptation target into a repository.

In one embodiment, the comparison operation converts each of the second adaptive targets stored in the information base into a comparative adaptive target, and when the comparison operation receives the new second adaptive targets again, the comparison operation will be new. The second adaptation target is compared with each comparison adaptation target to generate each second evaluation path.

In one of the embodiments, the comparison operation is a tabu search algorithm.

In one embodiment, the genetic algorithm performs selective replication, mating, and mutation operations on each of the first adaptive targets to generate the second adaptive target.

By the above, the present invention quickly calculates an optimal solution based on the pursuit of a single point search operation and a comparison operation based on the genetic pursuit, and can preserve the excellent individual and maintain the diversity of the group, thereby improving the operation speed.

Furthermore, the comparison operation of the present invention can store the excellent second evaluation path in the information base, so as to facilitate the calculation of the second evaluation path with the second evaluation path, so that the calculation time can be reduced and the calculation efficiency is improved. Calculation results.

For the convenience of the description, the central idea expressed by the present invention in the column of the above summary of the invention is expressed by the specific embodiments. Various items in the embodiments are depicted in terms of ratios, dimensions, amounts of deformation, or displacements that are suitable for illustration, and are not drawn to the proportions of actual elements, as set forth above.

Referring to FIG. 1 to FIG. 3, the present invention provides a hybrid genetic algorithm, which comprises the following steps:

Random step S1: receiving a pursuit target 10, randomly generating a complex resolution path 20 for the pursuit target 10. Further, in the embodiment of the present invention, for the pursuit target 10 to be solved, a feasible solution that may meet the pursuit target 10 is randomly generated, and each solution path 20 is also a feasible solution.

First conversion step S2: Convert each solution path 20 into a corresponding one of the first adaptation targets 30. Further, in the embodiment of the present invention, each solution path 20 needs to be converted into an adaptation function that can be used in different calculation manners used by different applications. Therefore, each first adaptation target 30 is equivalent to an adaptation function.

a first search step S3: comparing the first adaptation target 30 with the pursuit target 10, wherein, among the respective first adaptation targets 30, one of the first adaptation targets 30 meets the pursuit target 10, then Satisfying the condition, and meeting the first adaptation goal 30 is considered the best solution.

However, when any of the first adaptation targets 30 fails to satisfy the pursuit target 10, that is, when the infeasible solution is generated, the first adaptation target 30 is subjected to a genetic operation 40, and the genetic operation 40 is adapted to each first. The target 30 performs a selection copy 41, a mating 42 and a mutation 43 operation to generate a plurality of second adaptation targets 50. Wherein, each of the first adaptation targets 30 performs a selection 41 of the genetic operations 40, the purpose of which is to retain the first adaptation targets 30 that are more consistent with the pursuit target 10, and exclude the first adaptation targets 30 that do not completely meet the pursuit target 10 Excluded, for preliminary screening; the first adaptation target 30 that is more in line with the pursuit target 10 is subjected to the mating 42 of the genetic operation 40, and the first adaptation target 30 that is more in line with the pursuit target 10 is grouped by two. Performing a pairwise mating 42 produces a new first adaptation target 30, and whether mating 42 is determined by the mating probability, and the position after mating 42 is also randomly determined to maintain the diversity of each first adaptation target 30; Each of the first adaptation targets 30 generated 42 performs a mutation 43 of the genetic operation 40 to determine whether each of the first adaptation targets 30 is a mutation 43 in a random manner, and a plurality of second adaptation targets 50 are generated after the mutation 43.

The second searching step S4: performing a single-point search operation 60 on each of the second adaptive targets 50 to generate a plurality of first evaluation paths 61, wherein the single-point search operation 60 separates each of the second adaptive targets 50 and the pursuit targets 10 one by one. The first evaluation path 61 is generated to meet the pursuit target 10. In the embodiment of the present invention, the single point search operation 60 is a simulated annealing method, and the single point search operation 60 is an approximate solution. In a fixed time range, it is sought to compare each of the second adaptive targets 50 with the pursuit target 10 one by one in a large search range, thereby having the ability to jump off the regional minimum to quickly find the best solution.

The third searching step S5: performing a comparison operation 70 on each of the second adaptation targets 50 to generate a plurality of second evaluation paths 71, and in the calculation of the comparison operation 70, the comparison operation 70 stores the second adaptation targets 50. In the information library 72, each of the second adaptive targets 50 stored in the information base 72 is converted into a comparative adaptive target 73. During the continuous calculation, the comparison operation 70 continuously receives the new second adaptive targets again. 50. The comparison operation 70 compares each new second adaptation target 50 with each comparison adaptation target 73 to generate each second evaluation path 71. In the embodiment of the present invention, the information base 72 stores 10 comparisons. Adapt to goal 73.

Further, when the second adaptation target 50 is compared for the first time, the second adaptation target 50 is regarded as conforming to the pursuit target 10, and is converted into each comparison adaptation target 73; The third or more calculations, and when the second adaptation targets 50 generated by the first search step S3 are again entered into the comparison operation 70, the new second adaptation targets 50 are compared with each of the information bases 72. To adapt to the target 73 comparison, find the second adaptation target 50 that better matches the pursuit target 10, and then convert the more suitable second adaptation target 50 into the comparative adaptation target 73, and the original comparative adaptation target 73 will be replaced. In the embodiment of the present invention, the comparison operation 70 is a tabu search algorithm (Tabu Search), and the comparison operation 70 is an auxiliary heuristic rule, which uses the previous search results to avoid falling into a local optimal solution, thereby avoiding the roundabout search. Ensure effective exploration of diversity to ultimately achieve global optimization.

Furthermore, the comparison operation 70 can store the excellent second evaluation path 71 in the information base 72, so as to facilitate the calculation of the second evaluation path 71 with the second evaluation path, so that the calculation time can be reduced and the calculation efficiency can be improved. And calculation results.

In the embodiment of the present invention, the second searching step S4 and the third searching step S5 are performed simultaneously, and each of the second adaptive targets 50 and the pursuit target 10 are compared one by one by a single-point search operation 60, and the comparison operation 70 is performed. Avoid falling into the local optimal solution, so it can effectively shorten the calculation time and quickly converge to achieve the effect of fast calculation.

The second conversion step S6: converts each of the first evaluation paths 61 and each of the second evaluation paths 71 into a hybrid adaptive target 80. Further, the first evaluation path 61 corresponding to the pursuit target 10 is calculated by the single-point search operation 60, and the second evaluation paths 71 corresponding to the pursuit target 10 are calculated by the comparison operation 70, and the two are combined to find the most suitable pursuit. The path of goal 10, and then convert the most consistent path into mixed adaptation target 80.

Step S7 is obtained: the hybrid adaptation target 80 is compared with the pursuit target 10, and when the hybrid adaptation target 80 conforms to the pursuit target 10, the hybrid adaptation target 80 is regarded as an optimal solution.

For example: during the manufacturing process of a mechanical factory, a machine failure occurs, and a variable manufacturing schedule is generated. At this time, the pursuit of the production line schedule 10 is to minimize the original manufacturing schedule and the variation manufacturing schedule. The difference between the start of processing time and the completion of processing time.

First, the random step S1 is entered, and the solution target 10 is randomly generated to solve the solution paths 20 of the pursuit target 10, and each solution path 20 is how to allocate the processing time of the machine. Next, from the first conversion step S2, each of the solution paths 20 is converted into a corresponding first adaptation target 30 for the factory scheduling system for the factory scheduling system to perform operations.

Then, the converted first adaptive target 30 is compared with the pursuit target 10 by the first searching step S3. In each of the first adaptive targets 30, it is found whether there is a compliance to minimize the original manufacturing schedule and the variant manufacturing. The pursuit target 10 of the difference between the processing time and the completion of the processing time, if satisfied, the first adaptation target 30 is satisfied as the planning of the variation manufacturing schedule to solve the problem of the machine failure; if it is not satisfied A genetic operation 40 is performed to generate different second adaptation targets 50 to seek to otherwise satisfy the optimal solution for the pursuit target 10.

Then, each of the second adaptation targets 50 simultaneously performs the second search step S4 and the third search step S5, and compares each of the second adaptation targets 50 with the pursuit target 10 one by one by the single-point search operation 60, and simultaneously performs the comparison operation 70. Avoid falling into local best solutions and find manufacturing schedules that meet the target 10.

Then, the first evaluation path 61 and each of the second evaluation paths 71 are comprehensively judged to be converted into the hybrid adaptation target 80 by the second conversion step S6, and the manufacturing schedule that best matches the pursuit target 10 is found through the solution step S7, so that the variation is made. The start-up and end-of-process time of the manufacturing schedule minimizes variation from the original manufacturing schedule.

Thereby, the present invention quickly calculates the optimal solution based on the pursuit target 10 via the genetic operation 40 with the single-point search operation 60 and the comparison operation 70, and can preserve the excellent individuals and maintain the diversity of the group, thereby improving the operation speed.

The above embodiments are merely illustrative of the invention and are not intended to limit the scope of the invention. Any modifications or variations that are made without departing from the spirit of the invention are intended to be protected.

10‧‧‧ pursue the target

71‧‧‧Second assessment path

20‧‧‧Resolve the path

72‧‧ Information Library

30‧‧‧First adaptation target

73‧‧‧Compared to the target

40‧‧‧Genetic calculation

80‧‧‧ mixed adaptation goals

41‧‧‧Select copy

S1‧‧‧ random steps

42‧‧‧Matching

S2‧‧‧ first conversion step

43‧‧‧ Mutation

S3‧‧‧First search step

50‧‧‧ second adaptation target

S4‧‧‧Second search step

60‧‧‧Single point search operation

S5‧‧‧ third search step

61‧‧‧First assessment path

S6‧‧‧Second conversion step

70‧‧‧Comparative operation

S7‧‧‧Solution steps

1 is a schematic flow chart of the steps of the present invention. 2 is a schematic block diagram of the flow of the present invention. 3 is a schematic diagram of a comparison operation flow of the present invention.

Claims (7)

A hybrid genetic operation method, comprising the steps of: receiving a tracking target, randomly generating a complex solution path for the pursuit target; converting each of the solution paths into a corresponding one of the first adaptation targets; Comparing with the pursuit target, generating an infeasible solution, performing a genetic operation on the first adaptation target to generate a plurality of second adaptation targets; performing a single point search operation on the second adaptation target, Generating a plurality of first evaluation paths; performing a comparison operation on the second adaptation target to generate a plurality of second evaluation paths; converting each of the first evaluation paths and each of the second evaluation paths into a hybrid adaptation target; The hybrid adaptation target is compared with the pursuit target, and when the hybrid adaptation target meets the pursuit target, the hybrid adaptation target is regarded as an optimal solution. The hybrid genetic algorithm of claim 1, wherein the single-point search operation compares each of the second adaptive targets with the search target one by one to generate each of the first evaluation paths that meet the search target. The hybrid gene operation method according to claim 2, wherein the single point search operation is a simulated annealing method. The hybrid genetic algorithm of claim 1, wherein the comparing operation stores the second adaptation target into a repository. The hybrid gene operation method of claim 4, wherein the comparison operation converts each of the second adaptation targets stored in the information base into a comparison adaptation target, and when the comparison operation receives the new each When the target is adapted, the comparison operation compares each of the new second adaptation targets with each of the comparison adaptation targets to generate each of the second assessment paths. The hybrid gene operation method according to claim 5, wherein the comparison operation is a tabu search algorithm. The hybrid genetic algorithm according to claim 1, wherein the genetic operation performs selective copying, mating, and mutation operations on each of the first adaptive targets to generate the second adaptive target.