基于非支配排序的多目标PSO算法MATLAB实现

thinkfunny · 发表于 2020-10-22 15:39

EDA365欢迎您登录！

您需要登录才可以下载或查看，没有帐号？注册

x

本帖最后由 thinkfunny 于 2020-10-22 15:41 编辑

这一篇是Xue Bing在一区cybernetics发的论文，里面提出了两个多目标PSO特征选择算法，一个是NSPSO另一个是CMDPSO。其中NSPSO是参考了NSGA2的框架和思想。

伪代码

具体流程

①划分数据集为测试集和训练集
②初始化PSO算法
③迭代开始
④计算两个目标值（论文中是特征数和错误率）
⑤非支配排序
⑥拥挤距离度量并排序
⑥对每个粒子从第一前沿面选择一个粒子作为gbest，更新当前粒子
⑦调整粒子群
⑧迭代结束返回 B( N8 t" v" L% _" t4 y1 K9 n

MATLAB实现：
NSPSO：

注意其中FSKNN是我的问题的评价函数，包含两个目标值，都存入到pfitness中

function [solution,time,pop,pfitness,site,LeaderAVE] = NSPSO(train_F,train_L)
tic
global maxFES
dim = size(train_F,2);
FES = 1;
sizep = 30;
pop = rand(sizep,dim);
popv = rand(sizep,dim);
pfitness = zeros(sizep,2);
LeaderAVE = zeros(1,2);
while FES <maxFES
Off_P = zeros(sizep,dim);
Off_V = zeros(sizep,dim);
ofitness = zeros(sizep,2);
for i=1:sizep
[pfitness(i,1),pfitness(i,2)] = FSKNN(pop(i,: ),i,train_F,train_L);
end
Front = NDSort(pfitness(:,1:2),sizep);
[~,rank] = sortrows([Front',-CrowdingDistance(pfitness,Front)']);
LeaderSet = rank(1:10);
solution = pfitness(LeaderSet,: );
LeaderAVE(1) = mean(solution(:,1));
LeaderAVE(2) = mean(solution(:,2));
for i = 1:sizep
good = LeaderSet(randperm(length(LeaderSet),1));
r1 = rand(1,dim);
r2 = rand(1,dim);
Off_V(i,: ) = r1.*popv(i,: ) + r2.*(pop(good,: )-pop(i,: ));
Off_P(i,: ) = pop(i,: ) + Off_V(i,: );
end
for i=1:sizep
[ofitness(i,1),ofitness(i,2)] = FSKNN(Off_P(i,: ),i,train_F,train_L);
end
temppop = [pop;Off_P];
tempv = [popv;Off_V];
tempfiness = [pfitness;ofitness];
[FrontNO,MaxFNO] = NDSort(tempfiness(:,1:2),sizep);
Next = false(1,length(FrontNO));
Next(FrontNO<MaxFNO) = true;
PopObj = tempfiness;
fmax = max(PopObj(FrontNO==1,: ),[],1);
fmin = min(PopObj(FrontNO==1,: ),[],1);
PopObj = (PopObj-repmat(fmin,size(PopObj,1),1))./repmat(fmax-fmin,size(PopObj,1),1);
% Select the solutions in the last front
Last = find(FrontNO==MaxFNO);
del = Truncation(PopObj(Last,: ),length(Last)-sizep+sum(Next));
Next(Last(~del)) = true;
% Population for next generation
pop = temppop(Next,: );
popv = tempv(Next,: );
pfitness = tempfiness(Next,: );
fprintf('GEN: %2d Error: %.4f F:%.2f\n',FES,LeaderAVE(1),LeaderAVE(2));
FES = FES + 1;
end
[FrontNO,~] = NDSort(pfitness(:,1:2),sizep);
site = find(FrontNO==1);
solution = pfitness(site,: );
LeaderAVE(1) = mean(solution(:,1));
LeaderAVE(2) = mean(solution(:,2));
toc
time = toc;
end* u4 z, i, K( n9 t

拥挤距离代码：

function CrowdDis = CrowdingDistance(PopObj,FrontNO)
% Calculate the crowding distance of each solution front by front
[N,M] = size(PopObj);
CrowdDis = zeros(1,N);
Fronts = setdiff(unique(FrontNO),inf);
for f = 1 : length(Fronts)
Front = find(FrontNO==Fronts(f));
Fmax = max(PopObj(Front,: ),[],1);
Fmin = min(PopObj(Front,: ),[],1);
for i = 1 : M
[~,Rank] = sortrows(PopObj(Front,i));
CrowdDis(Front(Rank(1))) = inf;
CrowdDis(Front(Rank(end))) = inf;
for j = 2 : length(Front)-1
CrowdDis(Front(Rank(j))) = CrowdDis(Front(Rank(j)))+(PopObj(Front(Rank(j+1)),i)-PopObj(Front(Rank(j-1)),i))/(Fmax(i)-Fmin(i));
end
end
end
end
/ I% y% E. p; i2 h, p6 }( a7 P! n

Truncation.m代码：

function Del = Truncation(PopObj,K)
% Select part of the solutions by truncation
N = size(PopObj,1);
%% Truncation
Distance = pdist2(PopObj,PopObj);
Distance(logical(eye(length(Distance)))) = inf;
Del = false(1,N);
while sum(Del) < K
Remain = find(~Del);
Temp = sort(Distance(Remain,Remain),2);
[~,Rank] = sortrows(Temp);
Del(Remain(Rank(1))) = true;
end
end/ K* g# K9 B- a) o3 X

zmmdmn · 发表于 2020-10-26 10:57

NSPSO采用NSGA2的框架

帐号		自动登录	找回密码
密码			注册

基于非支配排序的多目标PSO算法MATLAB实现

EDA365欢迎您登录！

浏览过的版块

推荐内容 /1