1.背景介绍

遗传算法（Genetic Algorithm，简称GA）是一种基于生物进化理论的优化算法，由John Holland于20世纪70年代初提出。它通过模拟自然选择和遗传机制，利用群体中个体之间的遗传信息交流和变异来搜索问题的解空间。

遗传算法的设计灵感来源于达尔文的进化论。达尔文提出，自然界中的生物通过遗传信息的传递和变异，逐步适应环境并进化。类似地，遗传算法通过对问题解空间中的个体进行选择、交叉和变异操作，模拟了生物进化的过程，以寻找问题的最优解或次优解。

2.数学模型

遗传算法的核心思想是通过不断迭代的过程，从初始的随机个体群体出发，通过选择、交叉和变异操作产生新一代的个体群体，使得群体中的个体逐渐适应环境并优化问题的目标函数。具体而言，遗传算法的步骤如下：

初始化：随机生成初始的个体群体，代表解空间中的潜在解。
评估：根据问题的目标函数，对每个个体进行评估，计算其适应度值，表示解的优劣程度。
选择：根据适应度值，选择优秀的个体作为父代，用于产生下一代的个体。选择策略可以是轮盘赌选择、竞争选择等。
交叉：从父代个体中选取一对个体，通过交叉操作产生新的个体，将两个个体的染色体信息进行混合。
变异：对新生成的个体进行变异操作，引入随机性，增加搜索空间的多样性。
更新：用新生成的个体替换原有的个体群体，形成下一代个体群体。
终止条件：通过设定的终止条件，如达到最大迭代次数、目标函数值达到一定阈值等，判断算法是否停止。
返回最优解：遗传算法迭代完成后，返回适应度值最高的个体作为问题的解。

遗传算法具有全局搜索能力、自适应性和鲁棒性，适用于各种优化问题，尤其在复杂、多模态和高维的问题中表现出色。它在工程、运筹学、人工智能等领域都有广泛应用，并衍生出许多变种算法和改进方法，如遗传编程、进化策略等。

3.文件结构

在这里插入图片描述

crossover.m						% 交叉育种
elitism.m						% 精英化
GeneticAlgorithm.m				% 遗传算法
initialization.m				% 初始化
Main.m							% 主函数
mutation.m						% 变异
selection.m						% 选择
Sphere.m						%

4.详细代码及注释

4.1 crossover.m

function [child1 , child2] = crossover(parent1 , parent2, Pc, crossoverName)switch crossoverNamecase 'single'Gene_no = length(parent1.Gene);ub = Gene_no - 1;lb = 1;Cross_P = round (  (ub - lb) *rand() + lb  );Part1 = parent1.Gene(1:Cross_P);Part2 = parent2.Gene(Cross_P + 1 : Gene_no);child1.Gene = [Part1, Part2];Part1 = parent2.Gene(1:Cross_P);Part2 = parent1.Gene(Cross_P + 1 : Gene_no);child2.Gene = [Part1, Part2];case 'double'Gene_no = length(parent1);ub = length(parent1.Gene) - 1;lb = 1;Cross_P1 = round (  (ub - lb) *rand() + lb  );Cross_P2 = Cross_P1;while Cross_P2 == Cross_P1Cross_P2 = round (  (ub - lb) *rand() + lb  );endif Cross_P1 > Cross_P2temp =  Cross_P1;Cross_P1 =  Cross_P2;Cross_P2 = temp;endPart1 = parent1.Gene(1:Cross_P1);Part2 = parent2.Gene(Cross_P1 + 1 :Cross_P2);Part3 = parent1.Gene(Cross_P2+1:end);child1.Gene = [Part1 , Part2 , Part3];Part1 = parent2.Gene(1:Cross_P1);Part2 = parent1.Gene(Cross_P1 + 1 :Cross_P2);Part3 = parent2.Gene(Cross_P2+1:end);child2.Gene = [Part1 , Part2 , Part3];
endR1 = rand();if R1 <= Pcchild1 = child1;
elsechild1 = parent1;
endR2 = rand();if R2 <= Pcchild2 = child2;
elsechild2 = parent2;
endend

4.2 elitism.m

function [ newPopulation2 ] = elitism(population , newPopulation, Er)M = length(population.Chromosomes); % number of individuals 
Elite_no = round(M * Er);[max_val , indx] = sort([ population.Chromosomes(:).fitness ] , 'descend');% The elites from the previous population
for k = 1 : Elite_nonewPopulation2.Chromosomes(k).Gene  = population.Chromosomes(indx(k)).Gene;newPopulation2.Chromosomes(k).fitness  = population.Chromosomes(indx(k)).fitness;
end% The rest from the new population
for k = Elite_no + 1 :  MnewPopulation2.Chromosomes(k).Gene  = newPopulation.Chromosomes(k).Gene;newPopulation2.Chromosomes(k).fitness  = newPopulation.Chromosomes(k).fitness;
endend

4.3 GeneticAlgorithm.m

function [BestChrom]  = GeneticAlgorithm (M , N, MaxGen , Pc, Pm , Er , obj, visuailzation)cgcurve = zeros(1 , MaxGen);%%  Initialization
[ population ] = initialization(M, N);
for i = 1 : Mpopulation.Chromosomes(i).fitness = obj( population.Chromosomes(i).Gene(:) );
endg = 1;
disp(['Generation #' , num2str(g)]);
[max_val , indx] = sort([ population.Chromosomes(:).fitness ] , 'descend');
cgcurve(g) = population.Chromosomes(indx(1)).fitness;%% Main loop
for g = 2 : MaxGendisp(['Generation #' , num2str(g)]);% Calcualte the fitness valuesfor i = 1 : Mpopulation.Chromosomes(i).fitness = Sphere( population.Chromosomes(i).Gene(:) );endfor k = 1: 2: M% Selection[ parent1, parent2] = selection(population);% Crossover[child1 , child2] = crossover(parent1 , parent2, Pc, 'single');% Mutation[child1] = mutation(child1, Pm);[child2] = mutation(child2, Pm);newPopulation.Chromosomes(k).Gene = child1.Gene;newPopulation.Chromosomes(k+1).Gene = child2.Gene;endfor i = 1 : MnewPopulation.Chromosomes(i).fitness = obj( newPopulation.Chromosomes(i).Gene(:) );end% Elitism[ newPopulation ] = elitism(population, newPopulation, Er);cgcurve(g) = newPopulation.Chromosomes(1).fitness;population = newPopulation; % replace the previous population with the newly made
endBestChrom.Gene    = population.Chromosomes(1).Gene;
BestChrom.Fitness = population.Chromosomes(1).fitness;if visuailzation == 1plot( 1 : MaxGen , cgcurve);xlabel('Generation');ylabel('Fitness of the best elite')
endend

4.4 initialization.m

function [ population ] = initialization(M, N)for i = 1 : Mfor j = 1 : N population.Chromosomes(i).Gene(j) = [ round( rand() ) ];end
end

4.5 Main.m

clear
clc%% controling paramters of the GA algortihm
Problem.obj = @Sphere;
Problem.nVar = 20;M = 20; % number of chromosomes (cadinate solutions)
N = Problem.nVar;  % number of genes (variables)
MaxGen = 100;
Pc = 0.85;
Pm = 0.01;
Er = 0.05;
visualization = 1; % set to 0 if you do not want the convergence curve [BestChrom]  = GeneticAlgorithm (M , N, MaxGen , Pc, Pm , Er , Problem.obj , visualization)disp('The best chromosome found: ')
BestChrom.Gene
disp('The best fitness value: ')
BestChrom.Fitness

4.6 mutation.m

function [child] = mutation(child, Pm)Gene_no = length(child.Gene);for k = 1: Gene_noR = rand();if R < Pmchild.Gene(k) = ~ child.Gene(k);end
endend

4.7 selection.m

function [parent1, parent2] = selection(population)M = length(population.Chromosomes(:));if any([population.Chromosomes(:).fitness] < 0 ) % Fitness scaling in case of negative values scaled(f) = a * f + ba = 1;b = abs( min(  [population.Chromosomes(:).fitness] )  );Scaled_fitness = a *  [population.Chromosomes(:).fitness] + b;normalized_fitness = [Scaled_fitness] ./ sum([Scaled_fitness]);
elsenormalized_fitness = [population.Chromosomes(:).fitness] ./ sum([population.Chromosomes(:).fitness]);
end%normalized_fitness = [population.Chromosomes(:).fitness] ./ sum([population.Chromosomes(:).fitness]);[sorted_fintness_values , sorted_idx] = sort(normalized_fitness , 'descend');for i = 1 : length(population.Chromosomes)temp_population.Chromosomes(i).Gene = population.Chromosomes(sorted_idx(i)).Gene;temp_population.Chromosomes(i).fitness = population.Chromosomes(sorted_idx(i)).fitness;temp_population.Chromosomes(i).normalized_fitness = normalized_fitness(sorted_idx(i));
endcumsum = zeros(1 , M);for i = 1 : Mfor j = i : Mcumsum(i) = cumsum(i) + temp_population.Chromosomes(j).normalized_fitness;end
endR = rand(); % in [0,1]
parent1_idx = M;
for i = 1: length(cumsum)if R > cumsum(i)parent1_idx = i - 1;break;end
endparent2_idx = parent1_idx;
while_loop_stop = 0; % to break the while loop in rare cases where we keep getting the same index
while parent2_idx == parent1_idxwhile_loop_stop = while_loop_stop + 1;R = rand(); % in [0,1]if while_loop_stop > 20break;endfor i = 1: length(cumsum)if R > cumsum(i)parent2_idx = i - 1;break;endend
endparent1 = temp_population.Chromosomes(parent1_idx);
parent2 = temp_population.Chromosomes(parent2_idx);end

4.8 Sphere.m

function  [fitness_value] = Sphere( X )fitness_value = sum(X.^2);end

5.运行结果

在这里插入图片描述

6.参考文献

[1]Hayes-Roth F. Review of “Adaptation in Natural and Artificial Systems by John H. Holland”, The U. of Michigan Press, 1975[J]. ACM SIGART Bulletin,1975,53(53).