CN105978816B

CN105978816B - Multicast tree optimization method based on genetic framework

Info

Publication number: CN105978816B
Application number: CN201610265453.6A
Authority: CN
Inventors: 高超; 梁鸣心; 张自力
Original assignee: Southwest University
Current assignee: Southwest University
Priority date: 2016-04-27
Filing date: 2016-04-27
Publication date: 2020-11-13
Anticipated expiration: 2036-04-27
Also published as: CN105978816A

Abstract

A multicast tree optimization method based on genetic framework is used for solving the problem of how to fundamentally improve the local search capability of GA and keep the search range of the algorithm. The method comprises the following steps: s01, initializing parameters; s02, filtering the link; s03, initializing a population; s1, putting the chromosomes into a chromosome pool; s2, selecting a parent chromosome; s3, judging whether to execute an intersection operator for each pair of parent chromosomes, if so, turning to the step S4, otherwise, turning to the step S5; s4, reserving the multicast tree of the same link of the parent, and calculating the child multicast tree based on the reserved link; s5, judging whether a mutation operator is executed, if so, turning to the step S6, otherwise, turning to the step S7; s6, executing a mutation operator; s7, judging whether the evolution reaches the maximum propagation algebra, if so, turning to the step S8, otherwise, turning to the step S1; and S8, outputting the optimized multicast tree.

Description

Multicast tree optimization method based on genetic framework

[ technical field ] A method for producing a semiconductor device

The invention relates to the field of software algorithms, in particular to a multicast tree optimization method based on a genetic framework.

[ background of the invention ]

With the development of communication technology, the application range of multicast services is more and more extensive, and how to improve the Quality of Service (QoS) of a multicast network, balance network load, and optimize network resources becomes a key problem in current communication network research. The Multicast Tree Problem (MTP), which may be formulated as the Steiner Tree Problem (STP), is a typical NP-complete Problem. In conclusion, the MTP problem solving method has strong practical significance.

The solving algorithms of MTP can be divided into two main categories: a precision solution and an approximation solution. Common precision solutions include branch definition algorithms and dynamic programming algorithms. However, as the algorithm time complexity is too high, the calculation time increases in a geometric level along with the increase of the problem scale, and the algorithm is not feasible. Therefore, at present, more artificial intelligent optimization algorithms and improved algorithms thereof are used for solving, such as genetic algorithms, neural network algorithms, particle swarm optimization algorithms, ant colony optimization algorithms and the like.

Genetic Algorithm (GA), an optimization Algorithm proposed by professor John holland, university of Michigan, et al, to simulate the mechanism of biological evolution, has been successfully applied to solve for MTP. Although the method has better searching capability and higher solving quality, the defects of poor robustness, weak local searching capability and the like still limit the efficiency of the genetic algorithm.

In order to improve the efficiency of GA, researchers have proposed various improved algorithms. The current improvement starts from the following three aspects:

firstly, improving coding strategies, such as designing a gene coding table, so that genes in the coding table all represent links with better service quality, and the formed chromosome has better properties;

optimizing a crossover operator in the GA, for example, reserving the same link in a parent chromosome, and connecting the reserved links by a random path;

and thirdly, optimizing a mutation operator in the GA, so that the mutation operator has certain local search capability.

Although the local search capability of the algorithm can be improved to a certain extent by optimizing the coding tables and the intersection and mutation operators, the search range of the algorithm is reduced, so that the algorithm is easy to fall into the local optimal solution, and the optimization capability is limited. Therefore, how to fundamentally improve the local search capability of the GA and maintain the search range of the algorithm is one of the problems to be solved at present.

[ summary of the invention ]

The invention provides a multicast tree optimization method based on a genetic framework, which is used for solving the problems of how to fundamentally improve the local search capability of GA and keep the search range of an algorithm.

The invention relates to a multicast tree optimization method based on a genetic framework, which comprises the following steps: s01, initializing parameters; s02, filtering the link; s03, initializing a population; s1, putting the chromosomes into a chromosome pool; s2, selecting a parent chromosome; s3, judging whether to execute an intersection operator for each pair of parent chromosomes, if so, turning to the step S4, otherwise, turning to the step S5; s4, generating a child multicast tree based on the same link in the parent multicast tree; s5, judging whether a mutation operator is executed, if so, turning to the step S6, otherwise, turning to the step S7; s6, executing a mutation operator; s7, judging whether the evolution reaches the maximum propagation algebra, if so, turning to the step S8, otherwise, turning to the step S1; s8, outputting the optimized multicast tree; in step S4, generating child multicast tree, specifically, inputting network NG, node S, and node set DE having the same topology as the current topology into multiple nodesThe method comprises the following steps of obtaining a progeny multicast tree by using a mathematical model of the brevifilis: in the multi-head folliculus model, a node s is set as an inlet, and all points in a node set DE are outlets; by using

Representing the pressure value of the node i at the time t, the following equation can be obtained according to kirchhoff's law:

wherein D^t _ijRepresents the conductivity of the pipeline at the time t and is proportional to the diameter of the pipeline; l is_ijRepresenting edge e in NG_i，jThe weight of (2); m represents the number of target nodes; i is₀Representing the total flow in the network of Physarum polycephalum, I₀The value is always kept unchanged in the network evolution process; calculating the pressure value of each time node i according to the equation set

And calculating the pressure value of the node j at each moment according to the equation set

According to the relationship among fluid, conductivity and pressure in the pipeline described by Poiseue's law, the following equation can be obtained:

obtaining the flow Q in the pipeline at each moment according to the equation^t _ij(ii) a And then updating the conductivity of each pipeline in the multi-head folliculus vesiculosus network, which specifically comprises the following steps:

wherein

The conductivity of the pipeline at the time t +1 is represented, and after the conductivity of all the pipelines is updated, the maximum values of the conductivity change amounts of the pipelines at the time t +1 and the time t are calculated, which is specifically as follows:

when the value is less than beta, outputting a multicast tree NT composed of edges with conductivity greater than eta; otherwise, entering the iteration at the moment t + 1.

Wherein the initialized parameters in step S01 include: the population number N, the cross probability Pc, the variation probability Pm, the number Pn of offspring chromosomes generated in each generation and the maximum propagation algebra MG; initial pheromone quantity I in multi-headed Physarum velutinum network₀Value D of initial conductivity of each pipeline in a multiheaded Phycomycetes network⁰A conductivity change threshold β of PM convergence, a feedback equation coefficient k and a retention threshold η, and a retention link weight, and sets a source node s and a target node set DE.

In the initialization population in step S03, the specific process of each initial individual formation is as follows: setting a source node s as a current node c, namely c is s; randomly selecting one edge from adjacent edges of c, and marking as e_c，bIf b is not the destination node, updating the current node as b, repeating the operation until the destination node is accessed, and marking a subgraph formed by the accessed points and edges as T; randomly selecting one node from destination nodes which have not been visited as current node c, randomly selecting one edge from adjacent edges of c, and marking as e_c，bIf the b is not the node in the T, updating the current node as the b, and repeating the operation until the node in the T is accessed; and merging the accessed edge and point into T; if the destination node is not in T, the destination node which is not in T is randomly selected as the current node c again, and the steps are repeated until all the destination nodes are in T; finally, based on depth-first traversal from source node, deleting redundant edges to obtain initial multicast tree, and coding and forming initial multicast treeBecomes the original chromosome.

Specifically, in step S1, the newly generated chromosomes are placed in the chromosome pool, and the chromosomes in the current chromosome pool are sorted according to fitness, and N chromosomes with higher fitness are retained.

In step S2, a roulette selection method is used to select a Pn pair parent chromosome from the current chromosome pool.

In step S4, based on the same link in the parent multicast tree, specifically, a network NG having the same topology as the current topology is generated, and the weight of the edge in the NG is set according to the following formula:

wherein L is_ijRepresenting edge e in NG_i，jWeight of c_ij、

Representing cost and quality of service attributes, w, of the links connecting node i and node j in the route, respectively_nRepresents the importance of the corresponding quality of service in the current question; and then comparing the links in the parent chromosome, and setting the weight of the same links in the NG of the parent chromosome as the weight.

The specific process of executing the mutation operator in step S6 is as follows: randomly deleting one edge in the multicast tree to split the multicast tree into two connected components, wherein the connected component containing the node s is marked as A, and the other connected component is marked as B; randomly selecting a node from the B as a current node, and marking as c; randomly selecting a point from the neighbors of the current node c, marking the point as v, if v is in B, changing the current node into v, and reselecting the neighbors v of the current node until v is not in B; will be edge e_c，vAdding a multicast tree, and simultaneously changing the current node into v; detecting whether A and B in the multicast tree are combined into a connected component, if so, traversing and deleting redundant edges based on depth priority; otherwise, continuing to select neighbors based on the current node; and finally outputting the mutated chromosomes.

When it is determined in step S5 that no mutation operator is performed, after the propagation operation is completed on the parent chromosome by Pn, the fitness of the newly generated chromosome is calculated, and the new individual is added to the chromosome pool.

Compared with the prior art, the multicast tree optimization method based on the genetic framework assumes that a multi-head folliculus network corresponds to a Mobile Ad-hoc NETworks (MANET) when the MTP is solved. Nodes and pipelines in the multi-head plush bacteria network respectively correspond to routing nodes and links among the nodes in the mobile ad hoc network, and the length of the pipeline of the multi-head plush bacteria network corresponds to the weighted value of QoS attribute and cost of the links. Compared with the existing crossover operator, the optimization method has the following advantages: the local search capability of the crossover operator is improved, so that the crossover operator can obtain a local optimal solution in a small range by using a multiheaded folliculus mathematical model based on the gene of the parent chromosome. In MTP, the global optimal solution is also the local optimal solution at the same time, i.e. the local optimal solution is a necessary condition for the global optimal solution. Due to the enhancement of the local search capability, the crossover operator can generate better offspring individuals, and meanwhile, the search of a local solution space is completed by the execution of the crossover operator every time, so that the algorithm search range is enlarged, and the algorithm robustness is improved.

[ description of the drawings ]

FIG. 1 is a flow chart of an embodiment of the present invention, described in detail in the technical scheme;

FIG. 2 is a chart of an optimization algorithm solving MTP work machine, and the detailed description shows the beneficial effects compared with the prior art;

FIG. 3 is a schematic diagram of a crossover operator based on the mathematical model of the multi-headed Physarum. Wherein (a) and (b) represent selected parent multicast trees, and (c) represent correspondingly generated child multicast trees. (c) The bold sides in the middle represent the same links in the parent, and the dashed sides represent links selected by the multiheaded Phycomycetes mathematical model. See step 108 and step 109 in the technical scheme for a detailed description.

FIG. 4 is a diagram of MTP framework solution based on optimization genetic algorithm of a multiheaded Physarum mathematical model. For convenience of description, the optimized genetic algorithm is added with a prefix 'PM-' on the basis of the original algorithm, for example, the energy-saving genetic algorithm is marked as EEGA, and the optimized energy-saving genetic algorithm is collectively called as PM-EEGA.

FIG. 5, FIG. 6, FIG. 7 and FIG. 8 show the comparison results of the MTP performance obtained by the genetic algorithm optimized by the method of the present invention and the algorithm before the optimization.

The data set uses four networks generated by different network generators, denoted as D1, D2, D3, D4. Where D1, D2, and D3 are data sets of the minimum cost, delay, bandwidth limited multicast tree problem, and D4 is a data set of the minimum cost, delay, bandwidth, jitter limited multicast tree problem. To measure the effect of the algorithm, S is introduced_min、S_average、S_varianceThe index is specifically defined as follows:

(1) minimum value S_minRepresenting the cost or QoS attribute value of the optimal multicast tree obtained by the algorithm;

(2) average value S_averageSum variance S_varianceAfter the representative algorithm is repeatedly calculated for T times, the cost of the optimal multicast tree or the average value and the variance of the QoS attribute value are obtained. For example, S_averageIs calculated in a manner that

Wherein

And the attribute value of the optimal multicast tree calculated in the ith time is represented.

FIGS. 5, 6 and 7 compare the results of solving MTP under D1, D2 and D3 for PM-GAMAR, PM-EEGA, PM-ISGSA and GAMAR, EEGA and ISGSA, and show the cost and delayed S obtained by different algorithms_averageAnd S_varianceAnd (6) comparing. The results show that S of the post-optimization algorithm (PM-GA), regardless of cost or delay_averageAre significantly lower than the original algorithm (GA). S_averageThe reduction of the method indicates that the cross operator based on the multiheaded folliculus mathematical model effectively improves the optimizing capability of the genetic algorithm and improves the quality of the obtained multicast tree; at the same time S_varianceThe decrease in PM-GA indicates a greater improvement in the robustness of the PM-GA compared to GA.

FIG. 8 compares PM-EEGA, PM-ISGSA, EEGA, ISGSA and bee colony algorithms (Bees Life-based Algorithm (BLA), Bees Algorithm (BA), Marriage in honey bee Optimization Algorithm (MBO)) solve the result of MTP under D4. The result shows that compared with the bee colony algorithm, the PM-EEGA, the PM-ISGSA, the EEGA and the ISGSA can solve the MTP problem better, and the PM-EEGA and the PM-ISGSA are superior to the S of the EEGA and the ISGSA_averageAnd lower.

FIGS. 9 and 10 depict the S of cost and delay in solving the MTP problem at D2 for PM-EEGA, PM-ISGSA, EEGA, and ISGSA_averageAnd S_varianceAs the population breeds. The PM-EEGA and the PM-ISGSA have faster descending speed and earlier tend to be stable no matter the cost and the delay. And the cost and delay of convergence of PM-EEGA and PM-ISGSA compared to EEGA and ISGSA_averageAnd lower. At the same time, at S_varianceAlthough the PM-EEGA and the PM-ISGSA are increased in the comparison, the decrease speed of the PM-EEGA and the PM-ISGSA is higher, the final variance is lower, and the robustness is stronger.

In FIG. 9, at the beginning of the iteration, the S of PM-EEGA, PM-ISGSA and EEGA, ISGSA_averageThe difference is small. And the difference is gradually expanded along with the increase of the iteration steps, and finally the performances of the PM-EEGA and the PM-ISGSA are obviously superior to those of the EEGA and the ISGSA, which shows that the convergence speed and the optimization capability of the improved algorithms (the PM-EEGA and the PM-ISGSA) are better improved. Also, S is depicted in FIG. 10_varianceThe convergence condition of PM-EEGA and PM-ISGSA shows that the robustness is better. In addition, the cost and the delay show similarity in the iterative process, that is, the variation trends of (a) and (b) in fig. 9 and 10 are similar, which shows that the cross operator based on the multi-headed folliculus can better consider multiple attributes (such as cost and delay) and can handle more complex multi-QoS requirements.

[ detailed description ] embodiments

The embodiment provides a multicast tree optimization method based on a genetic framework, which is shown in fig. 1 and comprises the following main steps:

the flow starts at step 101.

In step 102, the main parameters are initialized: number of population N, crossover probability Pc, variationProbability Pm, the number Pn of offspring chromosomes generated in each generation, and the maximum propagation generation MG; initial pheromone quantity I in multi-headed Physarum velutinum network₀Value D of initial conductivity of each pipeline in a multiheaded Phycomycetes network⁰A conductivity change threshold β for PM convergence, a feedback equation coefficient k and a retention threshold η, and a retention link weight. And sets a source node s and a destination node set DE. According to experience, the specific values of the parameters are as follows: n200, Pc 0.9, Pm 0.05, Pn 180, MG 300, D⁰＝1，β＝10^-6，k＝2，η＝10^-3，＝10^-3。

In step 103, the links that do not meet the qos requirement are filtered according to the qos requirement of the multicast tree. Taking the typical QoS attribute bandwidth as an example, the bandwidth of the multicast tree is determined by the minimum value of the bandwidth of the links constituting the multicast tree. And traversing the known links and filtering the links with the bandwidth less than the minimum bandwidth limit.

At step 104, N chromosomes are randomly initialized and fitness is calculated to form a population of the initial generation. Taking the method of the initial individuals in the ISGSA as an example, the specific process of forming each initial individual is as follows:

first, the source node s is set as the current node c, i.e., c ═ s.

Randomly selecting one edge from adjacent edges of c, and marking as e_c，bIf b is not the destination node, i.e.

The current node is updated to b, i.e., c ═ b. And repeating the operation until the destination node is accessed. At this time, a sub-graph composed of the accessed point and edge is denoted as T.

Then randomly selecting a node from the destination nodes which have not been accessed as a current node c, randomly selecting an edge from the adjacent edges of c, and marking as e_c，bIf b is not the node in T, the current node is updated to be b, that is, c is equal to b. The above operation is repeated until a node in T is accessed. The accessed edge and point are then merged into T. At this time, if there are more destination nodes not in T, the destination nodes not in T are randomly selected again as the current node c until all destination nodes are in T.

And finally, based on depth-first traversal from the source node, deleting redundant edges to obtain an initial multicast tree, and coding the initial multicast tree to form an initial chromosome.

In step 105, the newly generated chromosomes are placed into a chromosome pool, the chromosomes in the current chromosome pool are sorted according to fitness, and only N chromosomes with higher fitness are reserved, namely, the chromosomes in the chromosome pool are updated and eliminated.

In step 106, Pn pairs of parent chromosomes are selected from the current chromosome pool by using a roulette selection method.

At step 107, it is determined for each pair of parent chromosomes whether to perform an intersection operator. If the determination result is "yes", the flow proceeds to step 108; otherwise, go to step 110.

In step 108, a network NG with the same topology as the current network is generated, that is, a network NG with the same topology as the current network is generated, and the weight of the edge in the NG is set according to the following formula:

L_ijrepresenting edge e in NG_i，jWeight of c_ij、

Representing the cost and quality of service attributes of the links connecting routes i and j, respectively, w_uRepresenting the degree of importance of the corresponding quality of service in the current problem.

And then comparing the links in the parent chromosome, and setting the weight of the same links in the NG of the parent chromosome as the weight.

In step 109, NG, s, DE is input into the multiheaded folliculus mathematical model to obtain the offspring multicast tree NT, which specifically comprises the following steps:

in the multi-head folliculus model, a node s is set as an inlet, and all points in a point set DE are outlets. By using

wherein D_ijRepresents the conductivity of the pipe, proportional to the diameter of the pipe; l is_ijRepresenting the weight of the edge in the NG; m represents the number of target nodes; i is₀Representing the total flow in the multi-head folliculus network, and the value is kept unchanged all the time in the network evolution process. The pressure value of the node i at each moment can be calculated according to the equation set

the flow Q in the pipeline at each moment can be obtained according to the equation^t _ij. And then updating the conductivity of each pipeline in the multi-head folliculus vesiculosus network, which specifically comprises the following steps:

after the conductivity of all the pipelines is updated, calculating the maximum value of the conductivity variation of the single pipeline at the time t +1 and the time t, specifically as follows:

In step 110, it is determined whether a mutation operator is performed. If the judgment result is yes, entering step 111; otherwise step 112 is entered.

In step 111, mutation operators are executed to generate mutated individuals. Taking the variation in the genetic algorithm ISGSA as an example, the specific process is as follows:

firstly, randomly deleting one edge in the multicast tree, splitting the multicast tree into two connected components, wherein the connected component containing the node s is marked as A, and the other connected component is marked as B.

And randomly selecting one node from the B as a current node, and marking as c.

And randomly selecting a point from the neighbors of the current node c, and marking the point as v. If v is in B, the current node is changed to v, i.e., c ═ v, and the neighbor v of the current node is reselected until v is not in B.

Then, the edge e is put_c，vAnd adding the multicast tree, and simultaneously changing the current node into v.

And detecting whether A and B are combined into a connected component in the multicast tree. If yes, traversing and deleting redundant edges based on depth priority; otherwise, continuing to select the neighbor based on the current node.

And finally outputting the mutated chromosomes.

At step 112, after Pn completes the propagation operation on the parent chromosomes, the fitness of the newly generated chromosomes is calculated, and the new individuals are added into the chromosome pool.

In step 113, it is determined whether the evolution has reached the maximum propagation passage number. If yes, go to step 114; otherwise, go back to step 105.

At step 114, the chromosome with the highest fitness is decoded and output as the optimal multicast tree.

The flow ends at step 115.

The description and use of the invention herein are illustrative and exemplary only, and are not intended to limit the scope of the invention to the embodiments described above. Variations and modifications of the embodiments disclosed herein are fully possible, and alternative and equivalent various components of the embodiments are well known to those skilled in the art. It will also be apparent to those skilled in the art that the present invention may be embodied in other forms, structures, arrangements, proportions, and with other components, materials, and parts, and that other modifications and variations of the embodiments disclosed herein, without departing from the spirit or essential characteristics thereof.

Claims

1. A multicast tree optimization method based on genetic framework is characterized by comprising the following steps:

s01, initializing parameters;

s02, filtering the link;

s03, initializing a population;

s1, putting the chromosomes into a chromosome pool;

s2, selecting a parent chromosome;

s3, judging whether to execute an intersection operator for each pair of parent chromosomes, if so, turning to the step S4, otherwise, turning to the step S5;

s4, generating a child multicast tree based on the same link in the parent multicast tree;

s5, judging whether a mutation operator is executed, if so, turning to the step S6, otherwise, turning to the step S7;

s6, executing a mutation operator;

s7, judging whether the evolution reaches the maximum propagation algebra, if so, turning to the step S8, otherwise, turning to the step S1;

s8, outputting the optimized multicast tree;

generating a child multicast tree in step S4, specifically, inputting a network NG, a node S, and a node set DE having the same topology as the current topology into a multi-headed folliculus mathematical model to obtain a child multicast tree, which includes the following steps:

in the multiheaded folliculus mathematical model, setting a node s as an inlet and all points in a node set DE as outlets; by using

And representing the pressure value of the node i, obtaining the following equation according to kirchhoff law:

wherein D^t _ijRepresents the conductivity of the pipeline at the time t and is proportional to the diameter of the pipeline; l is_ijRepresenting edge e in NG_i，jThe weight of (2); m represents the number of target nodes; i is₀Representing the initial total flow in the network of Physarum polycephalum, I₀The value is always kept unchanged in the network evolution process; the set of target nodes is DE; calculating the pressure value of each time node i according to the equation set

wherein

The method comprises the following steps of representing the conductivity of a pipeline at the t +1 moment, calculating the maximum value of the conductivity variation of the pipeline at the t +1 moment and the t moment after the conductivity of all the pipelines is updated, wherein the coefficient of a feedback equation is k, and the method comprises the following specific steps:

when this value is smaller than the conductivity change threshold β at which PM represents the mathematical model of the plurality of folliculorum polycephalum, or else an iteration at time t +1, the multicast tree NT consisting of edges whose conductivity is greater than the retention threshold η is output.

2. The genetic framework-based multicast tree optimization method according to claim 1, wherein the initialized parameters in step S01 include: the population number N, the number Pn of offspring chromosomes generated in each generation; initial total flow I in the multiheaded Phycomycetes network₀A conductivity change threshold β for PM convergence, a feedback equation coefficient k and a retention threshold η, and sets a source node s and a target node set DE.

3. The genetic-framework-based multicast tree optimization method according to claim 1, wherein in the initialization population in step S03, each initial individual is formed by the following specific process:

setting a source node s as a current node c, namely c is s;

randomly selecting one edge from adjacent edges of c, and marking as e_c，bIf b is not the destination node, updating the current node as b, repeating the operation until the destination node is accessed, and marking a subgraph formed by the accessed points and edges as T;

randomly selecting one node from destination nodes which have not been visited as current node c, randomly selecting one edge from adjacent edges of c, and marking as e_c，bIf the b is not the node in the T, updating the current node as the b, and repeating the operation until the node in the T is accessed; and merging the accessed edge and point into T;

if the destination node is not in T, the destination node which is not in T is randomly selected as the current node c again, and the steps are repeated until all the destination nodes are in T;

4. The genetic-framework-based multicast tree optimization method according to claim 1, wherein step S1 is to put the newly generated chromosomes into the chromosome pool, sort the chromosomes in the current chromosome pool according to fitness, and keep the N chromosomes with higher fitness.

5. The genetic-framework-based multicast tree optimization method of claim 1, wherein in step S2, Pn pairs of parent chromosomes are selected from the current chromosome pool by using a round-robin selection method.

6. The genetic-framework-based multicast tree optimization method of claim 1, wherein in step S4, a network NG having the same topology as the current topology is generated based on the same links in the parent multicast tree, and the weight of the edge in the NG is set according to the following formula:

wherein L is_ijRepresenting edge e in NG_i，jWeight of c_ijRepresenting the cost of connecting the links of node i and node j in the route,

represents a minimum cost service attribute;

a proxy delay service attribute;

an nth quality of service attribute representing a link of node i and node j; w is a₁Representative minimum cost garmentThe importance of the business attribute; w is a₂Representing the importance of the delayed service attribute; w is a_nRepresenting the importance of the nth quality of service attribute, and then comparing the links in the parent chromosome, and setting the weight of the same link in the NG of the parent chromosome as the reserved link weight.

7. The genetic framework-based multicast tree optimization method according to claim 1, wherein the mutation operator is executed in step S6 as follows:

randomly deleting one edge in the multicast tree to split the multicast tree into two connected components, wherein the connected component containing the node s is marked as A, and the other connected component is marked as B;

randomly selecting a node from the B as a current node, and marking as c;

randomly selecting a point from the neighbors of the current node c, marking the point as v, if v is in B, changing the current node into v, and reselecting the neighbors v of the current node until v is not in B;

will be edge e_c，vAdding a multicast tree, and simultaneously changing the current node into v;

detecting whether A and B in the multicast tree are combined into a connected component, if so, traversing and deleting redundant edges based on depth priority; otherwise, continuing to select neighbors based on the current node;

and finally outputting the mutated chromosomes.