智能优化算法-遗传算法（GA）

说明

基础内容源自 智能优化算法及其MATLAB实例（第二版）源程序

源程序为 matlab 代码，笔者在此基础上将 matlab 代码转化为 python 代码实现

一、标准遗传算法求函数极值

GA21.m

%%%%%%%%%%%%%%%%%%%%标准遗传算法求函数极值%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%初始化参数%%%%%%%%%%%%%%%%%%%%%%%%%%
clear all;              %清除所有变量
close all;              %清图
clc;                    %清屏
NP=50;                  %种群数量
L=20;                   %二进制数串长度
Pc=0.8;                 %交叉率
Pm=0.1;                 %变异率
G=100;                  %最大遗传代数
Xs=10;                  %上限
Xx=0;                   %下限
f=randint(NP,L);        %随机获得初始种群
% 现在比较新的 matlab 版本中，需要将 randint 函数换成 randi 函数
%%%%%%%%%%%%%%%%%%%%%%%%%遗传算法循环%%%%%%%%%%%%%%%%%%%%%%%%
for k=1:G
    %%%%%%%%%%%%将二进制解码为定义域范围内十进制%%%%%%%%%%%%%%
    for i=1:NP
        U=f(i,:);
        m=0;
        for j=1:L
            m=U(j)*2^(j-1)+m;
        end
        x(i)=Xx+m*(Xs-Xx)/(2^L-1);
        Fit(i)= func1(x(i));
    end
    maxFit=max(Fit);           %最大值
    minFit=min(Fit);           %最小值
    rr=find(Fit==maxFit);
    fBest=f(rr(1,1),:);        %历代最优个体   
    xBest=x(rr(1,1));
    Fit=(Fit-minFit)/(maxFit-minFit);  %归一化适应度值
    %%%%%%%%%%%%%%%%%%基于轮盘赌的复制操作%%%%%%%%%%%%%%%%%%%
    sum_Fit=sum(Fit);
    fitvalue=Fit./sum_Fit;
    fitvalue=cumsum(fitvalue);
    ms=sort(rand(NP,1));
    fiti=1;
    newi=1;
    while newi<=NP
        if (ms(newi))<fitvalue(fiti)
            nf(newi,:)=f(fiti,:);
            newi=newi+1;
        else
            fiti=fiti+1;
        end
    end   
    %%%%%%%%%%%%%%%%%%%%%%基于概率的交叉操作%%%%%%%%%%%%%%%%%%
    for i=1:2:NP
        p=rand;
        if p<Pc
            q=randint(1,L);
            for j=1:L
                if q(j)==1;
                    temp=nf(i+1,j);
                    nf(i+1,j)=nf(i,j);
                    nf(i,j)=temp;
                end
            end
        end
    end
    %%%%%%%%%%%%%%%%%%%基于概率的变异操作%%%%%%%%%%%%%%%%%%%%%%%
    i=1;
    while i<=round(NP*Pm)
        h=randint(1,1,[1,NP]);      %随机选取一个需要变异的染色体
        for j=1:round(L*Pm)         
            g=randint(1,1,[1,L]);   %随机需要变异的基因数
            nf(h,g)=~nf(h,g);
        end
        i=i+1;
    end
    f=nf;
    f(1,:)=fBest;                   %保留最优个体在新种群中
    trace(k)=maxFit;                %历代最优适应度
end
xBest;                              %最优个体
figure
plot(trace)
xlabel('迭代次数')
ylabel('目标函数值')
title('适应度进化曲线')

func1.m

%%%%%%%%%%%%%%%%%%%%%%%%%适应度函数%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function result=func1(x)
fit= x+10*sin(5*x)+7*cos(4*x);
result=fit;

GA21.py

import numpy as np
import matplotlib.pyplot as plt
import math

# 用来正常显示中文标签及负号
plt.rcParams['font.sans-serif']=['SimHei']
plt.rcParams['axes.unicode_minus'] = False 

# 定义目标函数 func1
def func1(x):
    return x+10*sin(5*x)+7*cos(4*x)  # func1.m 对应的适应度函数

# 初始化参数
NP = 50  # 种群数量
L = 20   # 二进制数串长度
Pc = 0.8 # 交叉率
Pm = 0.1 # 变异率
G = 100  # 最大遗传代数
Xs = 10  # 上限
Xx = 0   # 下限

# 随机获得初始种群
f = np.random.randint(2, size=(NP, L))   # 表示生成一个(NP, L)形状的二维数组,元素只能为0、1

# 遗传算法循环
trace = np.zeros(G)  # 创建一个全为0的数组
for k in range(G):
    # 将二进制解码为定义域范围内十进制
    x = np.zeros(NP)
    for i in range(NP):
        m = np.sum(f[i,:] * 2 ** np.arange(L)[::-1])   # 计算二进制数组f[i,:]的十进制等价值
        x[i] = Xx + m * (Xs - Xx) / (2**L - 1)    # 将十进制数映射到区间上下界
        Fit = np.array([func1(xi) for xi in x])   # 计算适应度值

    maxFit = np.max(Fit)           # 最大值
    minFit = np.min(Fit)           # 最小值
    rr = np.where(Fit == maxFit)[0]
    fBest = f[rr[0], :]            # 历代最优个体
    xBest = x[rr[0]]
    Fit = (Fit - minFit) / (maxFit - minFit)  # 归一化适应度值

    # 基于轮盘赌的复制操作
    sum_Fit = np.sum(Fit)
    fitvalue = Fit / sum_Fit
    fitvalue = np.cumsum(fitvalue)
    ms = np.sort(np.random.rand(NP))
    fiti = 0
    newi = 0
    nf = np.zeros((NP, L), dtype=int)
    while newi < NP:
        if ms[newi] < fitvalue[fiti]:
            nf[newi, :] = f[fiti, :]
            newi += 1
        else:
            fiti += 1

    # 基于概率的交叉操作
    for i in range(0, NP, 2):
        p = np.random.rand()
        if p < Pc:
            q = np.random.randint(1, L + 1)
            mask = np.random.randint(1, L + 1, size=L) == q
            nf[i, mask], nf[i + 1, mask] = nf[i + 1, mask], nf[i, mask]

    # 基于概率的变异操作
    i = 0
    while i <= round(NP * Pm):
        h = np.random.randint(1, NP + 1) - 1
        for _ in range(round(L * Pm)):
            g = np.random.randint(1, L + 1) - 1
            nf[h, g] = 1 - nf[h, g]
        i += 1

    f = nf
    f[0, :] = fBest  # 保留最优个体在新种群中
    trace[k] = maxFit  # 历代最优适应度

# 输出最优个体
print("xBest:", xBest)

# 绘图
plt.plot(trace)
plt.xlabel('迭代次数')
plt.ylabel('目标函数值')
plt.title('适应度进化曲线')
plt.show()

xBest: 7.8567865913263235