matlab pso svm,MATLAB 使用PSO优化LSSVM的程序

%% 清空环境

clc

clear

%%导入训练数据和测试数据

train = [3000 0.176 0.2 0.2 113.04;6000 0.181 0.2 0.2 226.08;9000 0.186 0.2 0.2 339.12;12000 0.193 0.2 0.2 452.16;

15000 0.198 0.2 0.2 565.2;18000 0.202 0.2 0.2 678.24;21000 0.208 0.2 0.2 791.28;24000 0.216 0.2 0.2 904.32;

22000 0.05 0.2 0.05 828.96;22000 0.15 0.2 0.15 828.96;22000 0.1 0.2 0.1 828.6 ;22000 0.2 0.2 0.2 828.96;

22000 0.25 0.2 0.25 828.96;22000 0.3 0.2 0.3 828.96;22000 0.35 0.2 0.35 828.96;22000 0.4 0.2 0.4 828.96;

22000 0.45 0.2 0.45 828.96];

train_out = [5.387 2.824 3.414 3.051 2.515 3.312 3.084 4.137 1.519 1.284 2.413 2.012 1.815 2.512 2.914 3.224 5.671];

test = [4500 0.179 0.2 0.2 169.56;7500 0.184 0.2 0.2 282.6;10500 0.189 0.2 0.2 395.64;13500 0.195 0.2 0.2 508.68;

16500 0.2 0.2 0.2 621.72;19500 0.205 0.2 0.2 734.76; 22500 0.212 0.2 0.2 847.8];

test_out = [3.8237 2.9526 3.3824 2.8132 2.1938 2.816 3.0887];

%%数据归一化

%%归一化方法1(利用libsvm工具箱函数归一化)

[train_data,test_data]=scaleForSVM(train,test,0,1)

[train_result,test_result,pstrain1]=scaleForSVM(train_out',test_out',0,1)

%%归一化方法2(利用svm自带mapminmax函数归一化)

%[train_data ,pstrain0] = mapminmax(train',0,1);

%[test_data] = mapminmax('apply',test',pstrain0);

%[train_result,pstrain1] = mapminmax(train_out,0,1);

%[test_result] = mapminmax('apply',test_out,pstrain1);

%train_data = train_data'

%train_result=train_result'

%test_data = test_data'

%% 参数初始化

%粒子群算法中的两个参数

c1 = 1.5; % c1 belongs to [0,2] c1:初始为1.5,pso参数局部搜索能力

c2 = 1.7; % c2 belongs to [0,2] c2:初始为1.7,pso参数全局搜索能力

maxgen=300; % 进化次数

sizepop=30; % 种群规模

popcmax=10^(3);% popcmax:初始为1000,SVM 参数c的变化的最大值.

popcmin=10^(-1);% popcmin:初始为0.1,SVM 参数c的变化的最小值.

popgmax=10^(2);% popgmax:初始为1000,SVM 参数g的变化的最大值

popgmin=10^(-2);% popgmin:初始为0.01,SVM 参数c的变化的最小值.

k = 0.5; % k belongs to [0.1,1.0];

Vcmax = k*popcmax;%参数 c 迭代速度最大值

Vcmin = -Vcmax ;

Vgmax = k*popgmax;%参数 g 迭代速度最大值

Vgmin = -Vgmax ;

eps = 10^(-8);

%%定义lssvm相关参数

type='f';

kernel = 'RBF_kernel';

proprecess='proprecess';

%% 产生初始粒子和速度

for i=1:sizepop

% 随机产生种群

pop(i,1) = (popcmax-popcmin)*rand(1,1)+popcmin ; % 初始种群

pop(i,2) = (popgmax-popgmin)*rand(1,1)+popgmin;

V(i,1)=Vcmax*rands(1,1); % 初始化速度

V(i,2)=Vgmax*rands(1,1);

% 计算初始适应度

gam=pop(i,1)

sig2=pop(i,2)

model=initlssvm(train_data,train_result,type,gam,sig2,kernel,proprecess);

model=trainlssvm(model);

%求出训练集和测试集的预测值

[train_predict_y,zt,model]=simlssvm(model,train_data);

[test_predict_y,zt,model]=simlssvm(model,test_data);

%预测数据反归一化

train_predict=mapminmax('reverse',train_predict_y',pstrain1)%训练集预测值

test_predict=mapminmax('reverse',test_predict_y',pstrain1) %测试集预测值

%计算均方差

trainmse=sum((train_predict-train_out).^2)/length(train_result)

testmse=sum((test_predict-test_out).^2)/length(test_result)

fitness(i)=testmse %以测试集的预测值计算的均方差为适应度值

end

% 找极值和极值点

[global_fitness bestindex]=min(fitness); % 全局极值

local_fitness=fitness; % 个体极值初始化

global_x=pop(bestindex,:); % 全局极值点

local_x=pop; % 个体极值点初始化

% 每一代种群的平均适应度

avgfitness_gen = zeros(1,maxgen);

tic

%% 迭代寻优

for i=1:maxgen

for j=1:sizepop

%速度更新

wV = 1; % wV best belongs to [0.8,1.2]为速率更新公式中速度前面的弹性系数

V(j,:) = wV*V(j,:) + c1*rand*(local_x(j,:) - pop(j,:)) + c2*rand*(global_x - pop(j,:));

if V(j,1) > Vcmax %以下几个不等式是为了限定速度在最大最小之间

V(j,1) = Vcmax;

end

if V(j,1) < Vcmin

V(j,1) = Vcmin;

end

if V(j,2) > Vgmax

V(j,2) = Vgmax;

end

if V(j,2) < Vgmin

V(j,2) = Vgmin; %以上几个不等式是为了限定速度在最大最小之间

end

%种群更新

wP = 1; % wP:初始为1,种群更新公式中速度前面的弹性系数

pop(j,:)=pop(j,:)+wP*V(j,:);

if pop(j,1) > popcmax %以下几个不等式是为了限定 c 在最大最小之间

pop(j,1) = popcmax;

end

if pop(j,1) < popcmin

pop(j,1) = popcmin;

end

if pop(j,2) > popgmax %以下几个不等式是为了限定 g 在最大最小之间

pop(j,2) = popgmax;

end

if pop(j,2) < popgmin

pop(j,2) = popgmin;

end

% 自适应粒子变异

if rand>0.5

k=ceil(2*rand);%ceil 是向离它最近的大整数圆整

if k == 1

pop(j,k) = (20-1)*rand+1;

end

if k == 2

pop(j,k) = (popgmax-popgmin)*rand+popgmin;

end

%%新粒子适应度值

gam=pop(j,1)

sig2=pop(j,2)

model=initlssvm(train_data,train_result,type,gam,sig2,kernel,proprecess)

model=trainlssvm(model)

%求出训练集和测试集的预测值

[train_predict_y,zt,model]=simlssvm(model,train_data)

[test_predict_y,zt,model]=simlssvm(model,test_data)

%预测数据反归一化

train_predict=mapminmax('reverse',train_predict_y',pstrain1)%训练集预测值

test_predict=mapminmax('reverse',test_predict_y',pstrain1) %测试集预测值

%计算均方差

trainmse=sum((train_predict-train_out).^2)/length(train_result)

testmse=sum((test_predict-test_out).^2)/length(test_result)

fitness(j)=testmse

end

%个体最优更新

if fitness(j) < local_fitness(j)

local_x(j,:) = pop(j,:);

local_fitness(j) = fitness(j);

end

if fitness(j) == local_fitness(j) && pop(j,1) < local_x(j,1)

local_x(j,:) = pop(j,:);

local_fitness(j) = fitness(j);

end

%群体最优更新

if fitness(j) < global_fitness

global_x = pop(j,:);

global_fitness = fitness(j);

end

if abs( fitness(j)-global_fitness )<=eps && pop(j,1) < global_x(1)

global_x = pop(j,:);

global_fitness = fitness(j);

end

end

fit_gen(i)=global_fitness;

avgfitness_gen(i) = sum(fitness)/sizepop;

end

toc

%% 结果分析

plot(fit_gen,'LineWidth',5);

title(['适应度曲线','(参数c1=',num2str(c1),',c2=',num2str(c2),',终止代数=',num2str(maxgen),')'],'FontSize',13);

xlabel('进化代数');ylabel('适应度');

bestc = global_x(1)

bestg = global_x(2)

gam=bestc

sig2=bestg

model=initlssvm(train_data,train_result,type,gam,sig2,kernel,proprecess)

model=trainlssvm(model)

%求出训练集和测试集的预测值

[train_predict_y,zt,model]=simlssvm(model,train_data)

[test_predict_y,zt,model]=simlssvm(model,test_data)

%预测数据反归一化

train_predict=mapminmax('reverse',train_predict_y',pstrain1)

test_predict=mapminmax('reverse',test_predict_y',pstrain1)

%计算均方差

trainmse=sum((train_predict-train_out).^2)/length(train_result)

testmse=sum((test_predict-test_out).^2)/length(test_result)

besttestmse=testmse