1 简介
信息化时代的快速发展，给人们的生活带来了很大的便捷。越来越多的人习惯在网上办事、网络购物、在线支付，俨然成为了一种常态。通过网络来获取各类信息也更加方便，这就给网络钓鱼带来了肥沃的生存土壤。网络钓鱼的不断泛滥，给人们的生产生活带来了巨大威胁，网民的财产安全受到侵犯，不利于经济的发展和社会的稳定。因此，加大网络钓鱼检测技术的研究成为当下的重点课题。目前网络钓鱼检测存在着效率不高、稳定性和泛化性较差等缺点，对于网络钓鱼检测技术的研究与改进变得尤为重要。本文在机器学习算法研究的基础上，通过灰狼算法对BP神经网络进行优化，并将其应用到钓鱼网站的检测上，取得了较好的效果。具体模型参靠底下文献。
2 部分代码
% Improved Grey Wolf Optimizer
function [Alpha_scores_CGWO,Alpha_pos_CGWO,Convergence_curve1]=GWO(SearchAgents_no,Max_iter,lb,ub,dim,objfun)
global kkk
%display('CGWO is optimizing your problem');
% SearchAgents_no=30;
% Max_iter=500;
% dim=10;
% objfun=@(x)sum(x.^2);
% ub=1; %/*lower bounds of the parameters. */
% lb=-1;%/*upper bound of the parameters.*/
%Alpha_scores_CGWO = zeros(1,runtime);
%for r=1:runtime
% initialize alpha, beta, and delta_pos
Alpha_pos_CGWO=zeros(1,dim);
Alpha_score=inf; %change this to -inf for maximization problems
Beta_pos=zeros(1,dim);
Beta_score=inf; %change this to -inf for maximization problems
Delta_pos=zeros(1,dim);
Delta_score=inf; %change this to -inf for maximization problems
%% Initialize the positions of search agents
% ub=ones(1,dim)*1; %/*lower bounds of the parameters. */
% lb=ones(1,dim)*(-1);%/*upper bound of the parameters.*/
% Range = repmat((ub-lb),[SearchAgents_no 1]);
% Lower = repmat(lb, [SearchAgents_no 1]);
% % 在搜索空间初始化食物源在-1和1之间
% Positions = rand(SearchAgents_no,dim) .* Range + Lower;
%% cat映射
% %灰狼位置混沌初始化种群
% x=double(zeros(1,dim)); % 定义初始序列
% X0=[0.1,0.3284]; % 初值
% y=x;
% z1=rand(1,dim);
% z2=rand(1,dim);
% % 赋初值
% z1(1)=X0(1);
% z2(1)=X0(2);
%
% for i=1:SearchAgents_no
% for j=1:dim
% x(i,j)=mod(z1(j)+z2(j),1); % 取余
% y(i,j)=mod(z1(j)+2*z2(j),1); % 取余
% z1(j)= x(i,j);
% z2(j)= y(i,j);
% end
% end
%
% X=y; % 利用y序列
%
% for i=1:SearchAgents_no
% for j=1:dim
% Positions_1(i,j) = lb+X(i,j)*(ub-lb);
% %Positions(i,j) = lb+X(i,j)*(ub-lb);
% end
% end
%% Logistic映射
%Logistic映射公式：x(t+1)=u*x(t)*(1-x(t))
% X1=double(zeros(1,dim)); % 定义初始序列
% z0 = rand(1,dim);
% u=4; % 混沌系统初始条件
% for i=1:SearchAgents_no
% for j=1:dim
% X1(i,j) = u*z0(j)*(1-z0(j));
% z0(j) = X1(i,j);
% end
% end
%
% for i=1:SearchAgents_no
% for j=1:dim
% Positions_1(i,j) = lb+X1(i,j)*(ub-lb);
% end
% end
%% Fuch映射
% X3=double(zeros(1,dim));
% z2=rand(1,dim);
% % 赋初值
% z2(1)=0.4835;
%
% for i=1:SearchAgents_no
% for j=1:dim
% X3(i,j) = cos(1/(z2(j)^2)); %产生混沌序列1
% z2(j) = X3(i,j);
% end
% end
%
% for i=1:SearchAgents_no
% for j=1:dim
% Positions_1(i,j) = lb+X3(i,j)*(ub-lb);
% end
% end
%% Tent映射
X2=double(zeros(1,dim));
z1=rand(1,dim);
% 赋初值
z1(1)=0.4835;
for i=1:SearchAgents_no
for j=1:dim
if z1(j)<=0.51 && z1(j)>=0
X2(i,j)=1/0.5*z1(j);
z1(j)= X2(i,j);
elseif z1(j)<=1 && z1(j)>0.5
X2(i,j)=1/0.5*(1-z1(j));
z1(j)= X2(i,j);
end
end
end
for i=1:SearchAgents_no
for j=1:dim
% Positions_1(i,j) = lb(j)+X2(i,j)*(ub(j)-lb(j));%求初始解x_i
Positions_1(i,j) = lb+X2(i,j)*(ub-lb);
end
end
%%
% 反向学习策略
min_number = min(Positions_1(i,:));
max_number = max(Positions_1(i,:));
for i=1:size(Positions_1,1)
for j=1:size(Positions_1,2)
Positions_2(i,j) = rand()*(min_number+max_number) - Positions_1(i,j);%每个初始解x_i对应的反向解OP_i
% Positions_2(i,j) = lb+ub - Positions_1(i,j);
end
end
Positions = [Positions_1;Positions_2];%两种解合并组成新的种群
for z =1: size(Positions,1)
fitness1(z,: )=feval(objfun,Positions(z,:));%求适应度函数
end
[m,index] = sort(fitness1);%适应度按升序排列
Positions = Positions(index(1:SearchAgents_no),:);%取前no最优的初始解得到新的灰狼初始种群
%% 同随机初始化位置对比
% 在搜索空间初始化食物源在-1和1之间
% ub1=ones(1,dim)*1; %/*lower bounds of the parameters. */
% lb1=ones(1,dim)*(-1);%/*upper bound of the parameters.*/
% Range1 = repmat((ub1-lb1),[SearchAgents_no 1]);
% Lower1 = repmat(lb1, [SearchAgents_no 1]);
% Positions_old = rand(SearchAgents_no,dim) .* Range1 + Lower1;
% figure
% h1 = plot(1:SearchAgents_no,x,'+','MarkerSize',4);
% xlabel('迭代次数','FontSize',10,'FontName','Times New Roman')
% ylabel('Cat混沌序列值','FontSize',10,'FontName','Times New Roman')
% 行数为40，列数为30
%此处的初始种群可以加以改进，比如混沌初始化，反向学习策略产生初始种群，已经有佳点集理论应用到这上面了。
l=1;% Loop counter
% Main loop
while l<=Max_iter
% 种群适应度和边界初始化
for i=1:size(Positions,1) % 其实是SearchAgents_no 种群个数 ，i 表示每一个灰狼个体
% Return back the search agents that go beyond the boundaries of the search space
Flag4ub=Positions(i,:)>ub;
Flag4lb=Positions(i,:)<lb;
Positions(i,:)=(Positions(i,:).*(~(Flag4ub+Flag4lb)))+ub.*Flag4ub+lb.*Flag4lb;
% Calculate objective function for each search agent
% 适应度函数
% fitness=fobj(Positions(i,:));
fitness = feval(objfun,Positions(i,:));
% Update Alpha, Beta, and Delta
% 更新 Alpha, Beta, and Delta 位置
if fitness<Alpha_score
Alpha_score=fitness; % Update alpha
Alpha_pos_CGWO=Positions(i,:);
end
fitness_Alpha = feval(objfun,Alpha_pos_CGWO); %fitness_Alpha的值是为了进行后面计算权重系数
% 关键是这个优先顺序
if fitness>Alpha_score && fitness<Beta_score
Beta_score=fitness; % Update beta
Beta_pos=Positions(i,:);
end
fitness_Beta = feval(objfun,Beta_pos); %同fitness_Alpha
if fitness>Alpha_score && fitness>Beta_score && fitness<Delta_score
Delta_score=fitness; % Update delta
Delta_pos=Positions(i,:);
end
fitness_Delta = feval(objfun,Delta_pos);%同fitness_Alpha
end
%% 利用Cat混沌局部搜索
min_X = min(Alpha_pos_CGWO);
max_X = max(Alpha_pos_CGWO);
for j=1:size(Alpha_pos_CGWO,2)
Z0(j) = (Alpha_pos_CGWO(j)-min_X)/(max_X-min_X);% 将最优位置映射到（0,1）之间
end
% 进行Tent映射迭代
M=double(zeros(1,dim)); % 定义初始序列
Z1=[0.1,0.4835]; % 初值
N=M;
z1=rand(1,dim);
z2=rand(1,dim);
% 赋初值
z1(1)=Z1(1);
z2(1)=Z1(2);
% 设置混沌迭代次数C_max
C_max=50;
for t=1:C_max
for j=1:size(Alpha_pos_CGWO,2)
M(t,j)=mod(z1(j)+z2(j),1); % 取余
N(t,j)=mod(z1(j)+2*z2(j),1); % 取余
z1(j)= M(t,j);
z2(j)= N(t,j);
end
end
V=N(t,:); % 得到第t次迭代混度搜索的序列（y序列）
% 下面开始进行载波到原解空间，产生最优解的邻域解
for j=1:size(Alpha_pos_CGWO,2)
%Alpha_pos_new(j) = Alpha_pos(j)+ (max_X-min_X)*(2*V(j)-1)/2;% 将最优位置映射到（0,1）之间
Alpha_pos_new(j) = Alpha_pos_CGWO(j)+ 3*(max_X-min_X)*(2*V(j)-1)/10;% 将最优位置映射到（0,1）之间
end
% 与原来的解相比较，保留最优解
if feval(objfun,Alpha_pos_new)>feval(objfun,Alpha_pos_CGWO)
Alpha_pos_CGWO=Alpha_pos_CGWO;
else
Alpha_pos_CGWO=Alpha_pos_new;
end
% 该算法考虑了群体的历史最优位置 &&保证了群体历史最优位置，并实现了狼群之间的交流，但并未考虑灰狼个体自身的经验
% 因此可以考虑引入PSO算法中的个体记忆功能
%a=2-l*((2)/Max_iter); % a decreases linearly fron 2 to 0
a=2*(1-(l/Max_iter)^2);
% Update the Position of search agents including omegas
for i=1:size(Positions,1) % 其实是SearchAgents_no 种群个数
for j=1:size(Positions,2) % 其实是变量维数 dim
r1=rand(); % r1 is a random number in [0,1] 产生一个数
r2=rand(); % r2 is a random number in [0,1]
% 此处可以改进
A1=2*a*r1-a; % Equation (3.3)
C1=2*r2; % Equation (3.4)
% 衡量灰狼与猎物之间的距离
D_alpha=abs(C1*Alpha_pos_CGWO(j)-Positions(i,j)); % Equation (3.5)-part 1
% 更新头狼位置
X1=Alpha_pos_CGWO(j)-A1*D_alpha; % Equation (3.6)-part 1
%更新副头狼位置
r1=rand();
r2=rand();
A2=2*a*r1-a; % Equation (3.3)
C2=2*r2; % Equation (3.4)
D_beta=abs(C2*Beta_pos(j)-Positions(i,j)); % Equation (3.5)-part 2
X2=Beta_pos(j)-A2*D_beta; % Equation (3.6)-part 2
% 更新普通狼位置
r1=rand();
r2=rand();
A3=2*a*r1-a; % Equation (3.3)
C3=2*r2; % Equation (3.4)
D_delta=abs(C3*Delta_pos(j)-Positions(i,j)); % Equation (3.5)-part 3
X3=Delta_pos(j)-A3*D_delta; % Equation (3.5)-part 3
% 可以采用三者的权值改进，不用1/3
w_Alpha = fitness_Alpha/(fitness_Alpha+fitness_Beta+fitness_Delta);
w_Beta = fitness_Beta/(fitness_Alpha+fitness_Beta+fitness_Delta);
w_Delta = fitness_Delta/(fitness_Alpha+fitness_Beta+fitness_Delta);
Positions(i,j)=w_Alpha*X1+w_Beta*X2+w_Delta*X3; % Equation (3.7)
%Positions(i,j)=(X1+X2+X3)/3;% Equation (3.7)
% 此处的灰狼最优位置更新可以结合人工蜂群算法的贪婪选择或者别的选择
%Positions(i,j) = Positions(i,j)+b*r3*(pBest(i,j)-Positions(i,j))+ (1-b)*rand()*( Alpha_pos_gbest(i,:)-Positions(i,j));
%Positions(i,j) = Positions(i,j)+b*rand()*(pBest(i,j)-Positions(i,j))+ (1-b)*rand()*( Alpha_pos_gbest(i,j)-Positions(i,j));
end % 每一只灰狼位置变量的更新完成
% r4=randn();
% Positions_new(i,:) = Positions(i,:)*(1+r4);
% % 优胜劣汰选择
% r5=rand();
% p = 0.1;
% if feval(objfun,Positions_new(i,:))>feval(objfun,Positions(i,:))&&r5<p
% Positions(i,:)=Positions(i,:);
% else
% Positions(i,:)=Positions_new(i,:);
% end
end % 该种群的所有灰狼位置更新
% 保存最优解的适应度
% Alpha_scores_CGWO(r,l)=Alpha_score;
Alpha_scores_CGWO(l)=Alpha_score;
% if mod(l,50)==0
% display(['At iteration ', num2str(l), ' the best fitness is ', num2str(Alpha_score)]);
% end
l=l+1;
Convergence_curve1(l)=Alpha_score;
end
% display([' CGWO 第 ', num2str(r), ' 次运行结束 ']);
% end
end
3 仿真结果
4 参考文献
[1]刘栋. (2019). 基于改进灰狼算法优化BP神经网络的研究和应用. (Doctoral dissertation, 安徽大学).
2303
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
