c_pso.m

粒子群算法

function [m_pattern]=C_PSO(m_pattern,patternNum)
        disType=DisSelDlg();%获得距离计算类型
        [centerNum iterNum]=InputClassDlg();%获得类中心数和最大迭代次数
        prticleNum=200;%初始化粒子数目
        %初始化中心和速度
        global Nwidth;
        for i=1:centerNum
            m_center(i).feature=zeros(Nwidth,Nwidth);
            m_center(i).prticleNum=0;
            m_center(i).index=i;
            m_velocity(i).feature=zeros(Nwidth,Nwidth);
       
         for i=1:prticleNum
             Particle(i).location=m_centerNum;%粒子各中心
             Particle(i).velocity=m_velocity;%粒子各中心速度
             Particle(i).fitness=0;%适应度
             P_id(i).location=m_center;%粒子最优中心
             P_id(i).velocity=m_velocity;%粒子最优速度
             P_id(i).fitness=0;%粒子最优适应度
         end
         P_gd.location=m_center;%全局粒子最优中心
         P_gd.velocity=m_velocity;%全局粒子最优速度
         P_gd.fitness=0;%粒子全局最优适应度
         P_gd.string=zeros(1,prticleNum);
         ptDitrib=zeros(prticleNum,prticleNum);%初始化粒子分布矩阵
      
         for i=1:prticleNum %生成随机粒子分布矩阵
             ptDitrib(i,:)=randperm(prticleNum);
             for j=1:prticleNum
                 if(ptDitrib(i,j)>centerNum)
                     ptDitrib(i,j)=fix(rand*centerNum+1);
                 end
             end
         end
       
        
       %生成初始粒子群
       for i=1:prticleNum
            for j=1:prticleNum
                m_pattern(j).category=ptDitrib(i,j);
            end
            for j=1:centerNum
                m_center(j)=CalCenter(m_center(j),m_pattern,patternNum);
            end
            Particle(i).locatoin=m_center;
       end
       
       %初始化参数
       w_max=1;
       w_min=0;
       h1=2;
       h2=2;
       
       
       for iter=1:iterNum
           %计算粒子适应度
           for i=1:patternNum
               temp=0;
               for j=1:patternNum
                   temp=temp+GetDistance(m_pattern(j),Particle(i).location(ptDitrib(i,j),disType));
               end
               if(temp==0) %最优解，直接退出
                   iter=iterNum+1;
                   break;
               end
               Particle(i).fitness=1/temp;
           end
           if(iter>iterNum)
               break;
           end
           w=w_max-iter*(w_max-w_min)/iterNum;%更新权重系数
           for i=1:particleNum %更新P_id,P_gd
               if(Particle(i).fitness>P_id(i).fitness)
                   P_id(i).fitness=Particle(i).fitness;
                   P_id(i).location=Particle(i).location;
                   P_id(i).velocity=Particle(i).velocity;
                   if(Particle(i).fitness>P_gd.fitness)
                       P_gd.fitness=Particle(i).fitness;
                       P_gd.location=Particle(i).location;
                       P_gd.velocity=Particle(i).velocity;
                       P_gd.string=ptDitrib(i,:);
                   end
               end
           end
           
           %更新粒子速度，位置
           for i=1:particleNum
               for j=1:centerNum
                   Particle(i).velocity(j).feature=w*Particle(i).velocity(j).feature
                   +h1*rand(Nwidth,Nwidth).*(P_id(i).location(j).feature-Particle(i).location(j).feature)
                   +h2*rand(Nwidth,Nwidth).*(P_gd.location(j).feature-Particle(i).location(j).feature);
               end
           end
           
           
           %最邻近聚类
           for i=1:particleNum
               for j=1:patternNum
                   min=inf;
                   for k=1:centerNum
                       tempDis=GetDistance(m_pattern(j),Particle(i).location(k),disType);
                       if(tempDis<min)
                           min=tempDis;
                           m_pattern(j).category=k;
                           ptDitrib(i,j)=k;
                       end
                   end
               end
               
               %重新计算聚类中心
               for j=1:centerNum
                   Particle(i).location(j)=CalCenter(Particle(i).location(j),m_parttern,patternNum);
               end
           end
           for i=1:partternNum
               m_pattern(i).category=P_gd.string(1,i);
           end
       end