基于迭代局部搜索和随机惯性权重的BA算法MATLAB实现（ILSSIWBA）

BA算法简介

经典蝙蝠算法MATLAB实现

该论文修改

作者在原有BA算法上进行3个修改

①跳出局部最优（扰动个体）

②使得算法变得稳定

③脉搏和响度修改，平衡局部与全局搜索

主要思想

流程图

MATLAB代码

function [bestMin, bestID] = ILSSIWBA()
%A new bat algorithm based on iterative local search and stochastic inertia weight
%omegaxyz.com QQ: 644327005
clc;
%% 经典BA参数设置
t = 1; 
maxT = 100; %最大迭代次数
dim = 30; %问题的维度
sizep = 50; %种群大小
xmin = -0.5;
xmax = 0.5; %位置向量的范围

A = 0.6.*ones(sizep,1);    % 响度 (不变或者减小)
r = zeros(sizep,1);      % 脉冲率 (不变或增加))
Qmin = 0;         % 最小频率
Qmax = 1;         % 最大频率

%% 初始化
Lb = xmin*ones(1,dim);
Ub = xmax*ones(1,dim);
pop = Lb+(Ub-Lb).*rand(sizep,dim); %种群初始化
popv = zeros(sizep,dim);   % 速度
Q = zeros(sizep,1);   % 频率

pfitness = zeros(dim,1);
for i = 1:sizep
    pfitness(i) = evaluate2(pop(i,:)); %评价
end
[bestMin, bestID]=min(pfitness);
bestS = pop(bestID, :);
bestArchive = zeros(maxT,1);

%% 论文中的新策略参数
umin = 0.4; %The second modification (SIW method)
umax = 0.9;
sigma = 0.2;

R0 = 0.1; %The 3th modification
Rupper = 0.7;
A0 = 0.9;
Alower = 0.6;

%% 具体迭代过程
while t <= maxT
    for i = 1:sizep
        Q(i)=Qmin+(Qmin-Qmax)*rand();
        w = umin +(umax-umin)*rand()+sigma*randn(); %The second modification (SIW method)
        popv(i,:)=w*popv(i,:)+(pop(i,:)-bestS)*Q(i); %The second modification (SIW method)
        Stemp = pop(i,:)+popv(i,:);
        % 脉冲率
         if rand>r(i)
             Stemp=bestS-1+2*rand(1,dim);
         end
         fitTemp = evaluate2(Stemp);
         if (fitTemp<=pfitness(i))&&(rand()<A(i))
            pop(i,:) = Stemp;
            pfitness(i) = fitTemp;
            A(i) = (A0-Alower)/(1-maxT)*(t-maxT)+Alower; %The 3th modification
            r(i) = (R0-Rupper)/(1-maxT)*(t-maxT)+Rupper; %The 3th modification
         end
         if fitTemp <= bestMin
            bestID = i;
            bestMin = fitTemp;
         	bestS = Stemp;
         end
    end
    
    interX = bestS.*rand(); %The 1th modification (ILS method)
    interXfit = evaluate2(interX);
    if interXfit < bestMin
        bestMin = interXfit;
        bestS = interX;
        pop(bestID,:) = bestS;
        pfitness(bestID,:) = bestMin;
    else
        if exp(bestMin-interXfit) > rand()
            bestMin = interXfit;
            bestS = interX;
            pop(bestID,:) = bestS;
            pfitness(bestID,:) = bestMin;
        end
    end
    
    bestArchive(t) = bestMin;
    fprintf('GEN: %d  min: %.4f\n', t, bestMin);
    t = t +1;
end

end%.m end

function [bestMin, bestID] = ILSSIWBA()

%A new bat algorithm based on iterative local search and stochastic inertia weight

%omegaxyz.com QQ: 644327005

clc;

%% 经典BA参数设置

t = 1;

maxT = 100; %最大迭代次数

dim = 30; %问题的维度

sizep = 50; %种群大小

xmin = -0.5;

xmax = 0.5; %位置向量的范围

A = 0.6.*ones(sizep,1); % 响度 (不变或者减小)

r = zeros(sizep,1); % 脉冲率 (不变或增加))

Qmin = 0; % 最小频率

Qmax = 1; % 最大频率

%% 初始化

Lb = xmin*ones(1,dim);

Ub = xmax*ones(1,dim);

pop = Lb+(Ub-Lb).*rand(sizep,dim); %种群初始化

popv = zeros(sizep,dim); % 速度

Q = zeros(sizep,1); % 频率

pfitness = zeros(dim,1);

for i = 1:sizep

pfitness(i) = evaluate2(pop(i,:)); %评价

end

[bestMin, bestID]=min(pfitness);

bestS = pop(bestID, :);

bestArchive = zeros(maxT,1);

%% 论文中的新策略参数

umin = 0.4; %The second modification (SIW method)

umax = 0.9;

sigma = 0.2;

R0 = 0.1; %The 3th modification

Rupper = 0.7;

A0 = 0.9;

Alower = 0.6;

%% 具体迭代过程

while t <= maxT

for i = 1:sizep

Q(i)=Qmin+(Qmin-Qmax)*rand();

w = umin +(umax-umin)*rand()+sigma*randn(); %The second modification (SIW method)

popv(i,:)=w*popv(i,:)+(pop(i,:)-bestS)*Q(i); %The second modification (SIW method)

Stemp = pop(i,:)+popv(i,:);

% 脉冲率

if rand>r(i)

Stemp=bestS-1+2*rand(1,dim);

end

fitTemp = evaluate2(Stemp);

if (fitTemp<=pfitness(i))&&(rand()<A(i))

pop(i,:) = Stemp;

pfitness(i) = fitTemp;

A(i) = (A0-Alower)/(1-maxT)*(t-maxT)+Alower; %The 3th modification

r(i) = (R0-Rupper)/(1-maxT)*(t-maxT)+Rupper; %The 3th modification

end

if fitTemp <= bestMin

bestID = i;

bestMin = fitTemp;

bestS = Stemp;

end

interX = bestS.*rand(); %The 1th modification (ILS method)

interXfit = evaluate2(interX);

if interXfit < bestMin

bestMin = interXfit;

bestS = interX;

pop(bestID,:) = bestS;

pfitness(bestID,:) = bestMin;

else

if exp(bestMin-interXfit) > rand()

bestMin = interXfit;

bestS = interX;

pop(bestID,:) = bestS;

pfitness(bestID,:) = bestMin;

end

bestArchive(t) = bestMin;

fprintf('GEN: %d min: %.4f\n', t, bestMin);

t = t +1;

end

end%.m end

评价函数：

function z = evaluate2(u)

%Weierstrass

dim= length(u);

sum_1=0;

sum_2=0;

sum_3=0;

a = 0.5;

b = 3;

for i=1:dim

for k=0:30

sum_1=sum_1+(a^k)*(cos(2*pi*(b^k)*(u(i)+0.5)));

end

sum_2=sum_2+sum_1;

sum_1 = 0;

end

for k=0:30

sum_3=sum_3+(a^k)*(cos(2*pi*(b^k)*0.5));

end

z=sum_2-dim*sum_3;

end

基于迭代局部搜索和随机惯性权重的BA算法MATLAB实现（ILSSIWBA）

BA算法简介

该论文修改

主要思想

流程图

MATLAB代码

大模型中的RepE表征工程

大模型也是一种优化器（LLM as Optimizer）

全栈开发与快速部署Demo

学术idea自动发现与生成

自回归语言模型（language model）Python实现

粉丝期待的三体电影宇宙（近四十部电影与电视剧集）

基于历史对比学习的时序知识图谱推理

泰拉瑞亚Terriaria快速部署Linux服务器

iPad生产力指南——编程

DeepScience：学术趋势预测与分析

2 评论

留下评论取消回复

BA算法简介

该论文修改

主要思想

流程图

MATLAB代码

相关文章

2 评论

留下评论取消回复