-
以下源代码为MOPSO的双目标规划,目标函数使用ZDT1来测试
-
多目标粒子群(MOPSO)
- 起源:1995年,受到鸟群觅食行为的规律性启发,James Kennedy和Russell Eberhart建立了一个简化算法模型,经过多年改进最终形成了粒子群优化算法(Particle Swarm Optimization, PSO) ,也可称为粒子群算法。粒子群算法的思想源于对鸟群觅食行为的研究,鸟群通过集体的信息共享使群体找到最优的目的地。
MATLAB程序:
clc; clear; close all; CostFunction = @(x) evaluate_objective(x); %目标函数ZDT1 nVar = 30; %变量个数 VarSize = [1 nVar]; %变量矩阵大小 VarMin = 0; %变量值定义域 VarMax = 1; %注意: 该函数变量不能出现负值 MaxIt = 200; %最大迭代次数 N = 40; %种群规模 nRep = 50; %档案库大小 w = 0.9; %惯性权重系数 wdamp = 0.99; %惯性权重衰减率 c1 = 1.7; %个体学习因子 c2 = 1.8; %全局学习因子 nGrid = 5; %每一维的分格数 alpha = 0.1; %膨胀率 beta = 2; %最佳选择压 gamma = 2; %删除选择压 mu = 0.1; %变异概率 empty_particle.Position = []; %粒子位置向量 empty_particle.Velocity = []; %粒子速度向量 empty_particle.Cost = []; %粒子目标值向量 empty_particle.Best.Position = []; %粒子最佳位置向量 empty_particle.Best.Cost = []; %粒子最佳目标值向量 empty_particle.IsDominated = []; %粒子被支配个体向量 empty_particle.GridIndex = []; %粒子栅格索引向量 empty_particle.GridSubIndex = []; %粒子栅格子索引向量 pop = repmat(empty_particle,N,1); %粒子初始空矩阵 for i = 1:N %初始化N个个体 % 产生服从均匀分布, VarSize大小的位置矩阵 pop(i).Position = unifrnd(VarMin,VarMax,VarSize); pop(i).Velocity = zeros(VarSize); pop(i).Cost = CostFunction(pop(i).Position); pop(i).Best.Position = pop(i).Position; pop(i).Best.Cost = pop(i).Cost; end pop = DetermineDomination(pop); rep = pop(~[pop.IsDominated]); Grid = CreateGrid(rep,nGrid,alpha); for i = 1:numel(rep) rep(i) = FindGridIndex(rep(i),Grid); % GridIndex = 绝对位置,.GridSubIndex = 坐标位置 end %MOPSO主循环 for it = 1:MaxIt for i = 1:N %逐一个体更新速度和位置,0.5的概率发生变异 leader = SelectLeader(rep,beta); %从支配个体轮盘赌选出全局最佳个体 rep = [rep;pop(~[pop.IsDominated])]; %添加新的最佳栅格位置到库 pop(i).Velocity = w*pop(i).Velocity + ... c1*rand(VarSize).*(pop(i).Best.Position-pop(i).Position)+ ... c2*rand(VarSize).*(leader.Position-pop(i).Position); %速度更新 pop(i).Position = pop(i).Position+pop(i).Velocity; %位置更新 pop(i).Position = limitToPosition(pop(i).Position,VarMin,VarMax); %限制变量变化范围 pop(i).Cost = CostFunction(pop(i).Position); %计算目标函数值 %应用变异策略 pm = (1-(it-1)/(MaxIt-1)^(1/mu)); % 变异概率逐渐变小 NewSol.Position = Mutate(pop(i).Position,pm,VarMin,VarMax); NewSol.Cost = CostFunction(NewSol.Position); % 计算变异后的目标值 if Dominates(NewSol,pop(i)) pop(i).Position = NewSol.Position; pop(i).Cost = NewSol.Cost; else %以0.5的概率决定是否接受变异 if rand < 0.5 pop(i).Position = NewSol.Position; pop(i).Cost = NewSol.Cost; end end if Dominates(pop(i),pop(i).Best) % 如果当前个体优于先前最佳个体,则替换之 pop(i).Best.Position = pop(i).Position; pop(i).Best.Cost = pop(i).Cost; else %以0.5的概率替换个体最佳 if rand <0.5 pop(i).Best.Position = pop(i).Position; pop(i).Best.Cost = pop(i).Cost; end end end %每个个体 rep = DetermineDomination(rep); rep = rep(~[rep.IsDominated]); Grid = CreateGrid(rep,nGrid,alpha); for i =1:numel(rep) rep(i) = FindGridIndex(rep(i),Grid); end if numel(rep) > nRep Extra = numel(rep)-nRep; for e = 1:Extra rep = DeleteOneRepMemebr(rep,gamma); end end disp(['迭代次数 =',num2str(it)]); w = w*wdamp; end figure(1); location = [rep.Cost]; %取最优结果 scatter(location(1,:),location(2,:),'filled','b'); xlabel('f1');ylabel('f2'); title('Pareto 最优边界图'); box on; %============================= %计算目标函数值 %============================= function f =evaluate_objective(x) f = []; f(1) = x(1); g = 1; sum = 0; V = size(x,2); for i = 2:V sum = sum + x(i); end sum = 9*(sum/(V-1)); g = g+sum; f(2) = g*(1-sqrt(x(1)/g)); f = [f(1);f(2)]; end %============================= %判断全局支配状况,返回0 = 非支配解 %============================= function pop =DetermineDomination(pop) nPop = numel(pop); for i =1:nPop pop(i).IsDominated = false; %初始化为互不支配 end for i = 1:nPop-1 for j = i+1:nPop if Dominates(pop(i),pop(j)) pop(j).IsDominated = true; end if Dominates(pop(j),pop(i)) pop(i).IsDominated = true; end end end end %============================= %判断两个目标值x,y的支配状态 % x支配y,返回1;y支配x,返回0 %============================= function b = Dominates(x,y) if isstruct(x) x=x.Cost; end if isstruct(y) y=y.Cost; end b=all(x<=y) && any(x<y); end %============================= %创建栅格矩阵 %============================= function Grid = CreateGrid(pop,nGrid,alpha) c = [pop.Cost]; cmin = min(c,[],2); cmax = max(c,[],2); dc = cmax-cmin; cmin = cmin-alpha*dc; cmax = cmax+alpha*dc; nObj = size(c,1); empty_grid.LB = []; empty_grid.UB = []; Grid = repmat(empty_grid,nObj,1); for j = 1:nObj cj = linspace(cmin(j),cmax(j),nGrid+1); Grid(j).LB = [-inf cj]; Grid(j).UB = [cj +inf]; end end %============================= %栅格索引定位 %============================= function particle = FindGridIndex(particle,Grid) nObj = numel(particle.Cost); nGrid = numel(Grid(1).LB); particle.GridSubIndex = zeros(1,nGrid); for j = 1:nObj particle.GridSubIndex(j) = find(particle.Cost(j)<=Grid(j).UB,1,'first'); %从左到右找到第一个目标值小于栅格值的位置 end particle.GridIndex = particle.GridSubIndex(1); for j = 2:nObj % 左上角开始数到右下角,先数行再换行继续数 particle.GridIndex = particle.GridIndex-1; particle.GridIndex = nGrid*particle.GridIndex; particle.GridIndex = particle.GridIndex + particle.GridSubIndex(j); end end %============================= %从全局支配个体中找出一个最佳个体 %============================= function leader = SelectLeader(rep,beta) GI = [rep.GridIndex]; OC = unique(GI); %一个栅格可能被多个支配解占用 N = zeros(size(OC)); for k =1:numel(OC) N(k) = numel(find(GI == OC(k))); end % 计算选择概率,为了增加多样性,尽量不选多次出现的个体 % 如果N大P就小, 即多次出现的栅格点被选中的概率小 P = exp(-beta*N); P = P/sum(P); sci = RouletteWheelSelection(P); %轮盘赌策略选择 sc = OC(sci); % 轮盘赌选择的栅格点 SCM = find(GI==sc); smi = randi([1 numel(SCM)]); sm = SCM(smi); leader = rep(sm); %当前全局最佳位置点 end %============================= %轮盘赌选择一个较好的支配个体 %============================= function i = RouletteWheelSelection(P) r = rand; C = cumsum(P); i = find(r<=C,1,'first'); end %============================= %限制变量变化范围在定义域内 %============================= function Position = limitToPosition(Position,VarMin,VarMax) for i =1:size(Position,2) if Position(i)<VarMin Position(i) = VarMin; elseif Position(i) > VarMax Position(i) = VarMax; end end end %============================= %删除档案库中的一个个体 %============================= function rep = DeleteOneRepMemebr(rep,gamma) GI = [rep.GridIndex]; OC = unique(GI); N = zeros(size(OC)); for k = 1:numel(OC) N(k) = numel(find(GI == OC(k))); end P = exp(gamma*N); P = P/sum(P); sci = RouletteWheelSelection(P); sc = OC(sci); SCM = find(GI == sc); smi = randi([1 numel(SCM)]); sm = SCM(smi); rep(sm) = []; end %============================= %使用变异策略 %============================= function xnew = Mutate(x,pm,VarMin,VarMax) nVar = numel(x); j = randi([1 nVar]); dx = pm*(VarMax-VarMin); lb = x(j)-dx; if lb<VarMin lb=VarMin; end ub = x(j)+dx; if ub > VarMax ub = VarMax; end xnew = x; xnew(j) = unifrnd(lb,ub); end
帕累托最优边界: