-
Notifications
You must be signed in to change notification settings - Fork 41
/
RBFPSO.m
178 lines (167 loc) · 5.27 KB
/
RBFPSO.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
clc;
clear;
tic;
SamNum=48; %训练样本数
TargetSamNum=3; %测试样本数
InDim=1; %样本输入维数
UnitNum=2; %隐节点数
MaxEpoch=1200; %最大训练次数
num=2;%对应四个特征
%E0=0.2; %目标误差
gbesthistory=[];
% 根据目标函数获得样本输入输出（训练样本）
rand('state',sum(100*clock));
%NoiseVar=0.0005;
%Noise=NoiseVar*randn(1,SamNum);
load data
%归一化
data1=data';
data=mapminmax(data1,0,1);
data=data';
%建立训练集测试集
x_train=[data(1:48,1).';data(1:48,2).';data(1:48,3).';data(1:48,4).'];
x_test=[data(2:49,num).'];
y_train=[data(49:51,1).';data(49:51,2).';data(49:51,3).';data(49:51,4).'];
y_test=[data(50:52,num).'];
SamIn=x_train;
SamOut=x_test;
%测试样本
TargetIn=y_train;
TargetOut=y_test;
%粒子群算法中的两个参数
c1 = 1.49445;
c2 = 1.49445;
popcount=10; %粒子数
poplength=6; %粒子维数
Wstart=0.9;%初始惯性权值
Wend=0.2;%迭代次数最大时惯性权值
%个体和速度最大最小值
Vmax=1;
Vmin=-1;
popmax=4;
popmin=-4;
%粒子位置速度和最优值初始化
for i=1:popcount
pop(i,:)=rand(1,9);%初始化粒子位置
V(i,:)=rand(1,9);%初始化粒子速度
%计算粒子适应度值
Center=pop(i,1:3);
SP=pop(i,4:6);
W=pop(i,7:9);
Distance=dist(Center',SamIn);
SPMat=repmat(SP',1,SamNum);%repmat具体作用
UnitOut=radbas(Distance./SPMat);%径向基函数
NetOut=W*UnitOut;%网络输出
Error=SamOut-NetOut;%网络误差
%SSE=sumsqr(Error);
%fitness(i)=SSE;
RMSE=sqrt(sumsqr(Error)/SamNum);
fitness(i)=RMSE;
%fitness(i)=fun(pop(i,:));
end
%适应度函数（适应度值为RBF网络均方差）
[bestfitness bestindex]=min(fitness);
gbest=pop(bestindex,:);%全局最优值
pbest=pop;%个体最优值
pbestfitness=fitness;%个体最优适应度值
gbestfitness=bestfitness;%全局最优适应度值
%迭代寻优
for i=1:MaxEpoch
Vmax=1.00014^(-i);
Vmin=-1.00014^(-i);
for j=1:popcount
% if (fitness(j)<gbestfitness|fitness==gbestfitness)
% S(j)=0;
%end
%S(j)=1-(fitness(j)/100)^2;
% GW(j)=Wstart-S(j)*(Wstart-Wend);
% GW(j)=Wend+(GW(j)-Wend)*(MaxEpoch-i)/MaxEpoch;
GW=Wstart-(Wstart-Wend)*i/MaxEpoch;
%速度更新(第一种方法精度最高)
V(j,:) = 1.000009^(-i)*(gbestfitness/fitness(j)+2)*rand*V(j,:) + c1*rand*(pbest(j,:) - pop(j,:)) + c2*rand*(gbest - pop(j,:));
%V(j,:) = GW*((fitness(j)/2000)^2+1)*rand*V(j,:) + c1*rand*(pbest(j,:) - pop(j,:)) + c2*rand*(gbest - pop(j,:));
%V(j,:) = GW*V(j,:) + c1*rand*(pbest(j,:) - pop(j,:)) + c2*rand*(gbest - pop(j,:));
%V(j,:) = 0.9*V(j ,:) + c1*rand*(pbest(j,:) - pop(j,:)) + c2*rand*(gbest - pop(j,:));
%V(j,:) = 0.9*1.0003^(-j)* V(j,:) + c1*rand*(pbest(j,:) - pop(j,:)) + c2*rand*(gbest - pop(j,:));
%V(j,:) = (gbestfitness/(exp(-fitness(j))+1)+0.5)*rand*V(j,:) + c1*rand*(pbest(j,:) - pop(j,:)) + c2*rand*(gbest - pop(j,:));
V(j,find(V(j,:)>Vmax))=Vmax;
V(j,find(V(j,:)<Vmin))=Vmin;
%粒子更新
pop(j,:)=pop(j,:)+0.5*V(j,:);
pop(j,find(pop(j,:)>popmax))=popmax;
pop(j,find(pop(j,:)<popmin))=popmin;
%计算粒子适应度值
Center=pop(j,1:3);
SP=pop(j,4:6);
W=pop(j,7:9);
Distance=dist(Center',SamIn);
SPMat=repmat(SP',1,SamNum);%repmat具体作用
UnitOut=radbas(Distance./SPMat);
NetOut=W*UnitOut;%网络输出
Error=SamOut-NetOut;%网络误差
%SSE=sumsqr(Error);
%fitness(j)=SSE;
RMSE=(sumsqr(Error)/SamNum);
fitness(j)=RMSE;
% Center=pop(j,1:10);
% SP=pop(j,11:20);
% W=pop(j,21:30);
% fitness(j)=fun(pop(j,:));
end
for j=1:popcount
%个体最优更新
if fitness(j) < pbestfitness(j)
pbest(j,:) = pop(j,:);
pbestfitness(j) = fitness(j);
end
%群体最优更新
if fitness(j) < gbestfitness
gbest = pop(j,:);
gbestfitness = fitness(j);
end
end
gbesthistory=[gbesthistory,gbest];
%mse(i)=gbestfitness;
%将群体最优值赋给RBF参数
Center=gbest(1,1:3);
SP=gbest(1,4:6);
W=gbest(1,7:9);
%Center=gbest(1,1:5);
%SP=gbest(1,11:20);
% W=gbest(1,21:30);
Distance=dist(Center',SamIn);
SPMat=repmat(SP',1,SamNum);%repmat具体作用
UnitOut=radbas(Distance./SPMat);
NetOut=W*UnitOut;%网络输出
Error=SamOut-NetOut;%网络误差
%sse(i)=sumsqr(Error);
mse(i)=(sumsqr(Error)/SamNum);
% sse(i)=fun(gbest);
%if sse(i)<E0,break,end
end
toc;
% 测试
Center=gbest(1,1:3);
SP=gbest(1,4:6);
W=gbest(1,7:9);
TestDistance=dist(Center',TargetIn);
TesatSpreadsMat=repmat(SP',1,TargetSamNum);
TestHiddenUnitOut=radbas(TestDistance./TesatSpreadsMat);
TestNNOut=W*TestHiddenUnitOut;
%作图 分别在训练集和测试集上
subplot(1,2,1)
plot(1:length(NetOut),NetOut,'*',1:length(NetOut),SamOut,'o')
title('In Train data')
subplot(1,2,2)
plot(1:3,TestNNOut,'*',1:3,TargetOut,'o')
title('In Test data')
%求出误差 训练集和测试集
train_error=sum(abs(SamOut-NetOut))/length(SamOut);
test_error=sum(abs(TargetOut-TestNNOut))/length(TargetOut);
% 绘制学习误差曲线
figure
hold on
grid
%[xx,Num]=size(errhistory);
plot(mse,'k-');
legend('mse')