【SVM分类】基于灰狼算法优化支持向量机实现数据分类matlab代码-CFANZ编程社区

1 简介

近年来，已有越来越多的建模方法被相关学者提出用来解决分类识别、风险预测、效能评估等问题，这些建模方法包括：时间序列分析、灰色理论、神经网络等。但是时间序列分析，方法复杂且预测精度较低；灰色理论需要规律性的数据；神经网络方法易出现过拟合以及易陷入局部极值等问题。支持向量机（（Support Vector Machine，SVM）是一种基于结构风险最小化且有着强大的泛化能力的建模方法。它可以很好地解决小样本、非线性，以及陷入局部极值等问题。然而，SVM的学习能力和泛化能力取决于合适的参数选择，这些参数直接影响了模型的性能。因此，近年来越来越多的参数优化方法被应用于 SVM的参数选择问题，比如网格搜索法、粒子群算法、蝙蝠算法等，然而网格搜索法运算量大，搜索效率低；粒子群算法和蝙蝠算法在参数寻优过程中会出现收敛速度慢、易陷入局部极值的问题。本文提出的 GWO—SVM模型，仿真结果表明，该模型预测精度较高。

2 部分代码

clc
clear all
close all
n=30;           % Population size, typically 10 to 25
p=0.8;           % probabibility switch

% Iteration parameters
N_iter=3000;            % Total number of iterations
fitnessMSE = ones(1,N_iter);

% % Dimension of the search variables Example 1
d=2;
Lb = -1*ones(1,d);
Ub = 1*ones(1,d);


% % Dimension of the search variables Example 2
% d=3;
% Lb = [-2 -1 -1];
% Ub = [2 1 1];

%
% % Dimension of the search variables Example 3
% d=3;
% Lb = [-1 -1 -1];
% Ub = [1 1 1];
%
%
% % % Dimension of the search variables Example 4
% d=9;
% Lb = -1.5*ones(1,d);
% Ub = 1.5*ones(1,d);

% Initialize the population/solutions
for i=1:n,
   Sol(i,:)=Lb+(Ub-Lb).*rand(1,d);
   % To simulate the filters use fitnessX() functions in the next line
   Fitness(i)=fitness(Sol(i,:));
end

% Find the current best
[fmin,I]=min(Fitness);
best=Sol(I,:);
S=Sol;

% Start the iterations -- Flower Algorithm
for t=1:N_iter,
   % Loop over all bats/solutions
   for i=1:n,
       % Pollens are carried by insects and thus can move in
       % large scale, large distance.
       % This L should replace by Levy flights
       % Formula: x_i^{t+1}=x_i^t+ L (x_i^t-gbest)
       if rand>p,
           %% L=rand;
           L=Levy(d);
           dS=L.*(Sol(i,:)-best);
           S(i,:)=Sol(i,:)+dS;
           
           % Check if the simple limits/bounds are OK
           S(i,:)=simplebounds(S(i,:),Lb,Ub);
           
           % If not, then local pollenation of neighbor flowers
       else
           epsilon=rand;
           % Find random flowers in the neighbourhood
           JK=randperm(n);
           % As they are random, the first two entries also random
           % If the flower are the same or similar species, then
           % they can be pollenated, otherwise, no action.
           % Formula: x_i^{t+1}+epsilon*(x_j^t-x_k^t)
           S(i,:)=S(i,:)+epsilon*(Sol(JK(1),:)-Sol(JK(2),:));
           % Check if the simple limits/bounds are OK
           S(i,:)=simplebounds(S(i,:),Lb,Ub);
       end
       
       % Evaluate new solutions
       % To simulate the filters use fitnessX() functions in the next
       % line
       Fnew=fitness(S(i,:));
       % If fitness improves (better solutions found), update then
       if (Fnew<=Fitness(i)),
           Sol(i,:)=S(i,:);
           Fitness(i)=Fnew;
       end
       
       % Update the current global best
       if Fnew<=fmin,
           best=S(i,:)   ;
           fmin=Fnew   ;
       end
   end
   % Display results every 100 iterations
   if round(t/100)==t/100,
       best
       fmin
   end
   
   fitnessMSE(t) = fmin;
   
end
%figure, plot(1:N_iter,fitnessMSE);
% Output/display
disp(['Total number of evaluations: ',num2str(N_iter*n)]);
disp(['Best solution=',num2str(best),'   fmin=',num2str(fmin)]);
figure(1)
plot( fitnessMSE)
xlabel('Iteration');
ylabel('Best score obtained so far');