MATLAB神经网络PSO-BP训练代码解析

function main
% This is a FNN trained by PSO and BP. After trained by PSO, the NN is trained by BP
clc;
clear all;
MaxRunningTime=10;        
MaxHiddenLayerNode=10;    
%根据目标函数获得样本输入输出
rand('state',sum(100*clock));
TrainSamIn=-4:0.08:4;
TrainSamOut=1.1*(1-TrainSamIn+2*TrainSamIn.^2).*exp(-TrainSamIn.^2/2);
TestSamIn=-3.96:0.08:3.96;
TestSamOut=1.1*(1-TestSamIn+2*TestSamIn.^2).*exp(-TestSamIn.^2/2);
[xxx,TrainSamNum]=size(TrainSamIn);
[xxx,TestSamNum]=size(TestSamIn);
%
for HiddenUnitNum=3:MaxHiddenLayerNode
    fprintf('\n the hidden layer node ');HiddenUnitNum
    AllTrainNNOut=[];
    AllTestNNOut=[];
    for t=1:MaxRunningTime
        fprintf('the current running times is');t
        [NewW1,NewB1,NewW2,NewB2]=PSOTrain(TrainSamIn,TrainSamOut,HiddenUnitNum);
        TrainNNOut=PSONetwork(NewW1,NewB1,NewW2,NewB2,TrainSamIn);
        AllTrainNNOut=[AllTrainNNOut;TrainNNOut];
        TestNNOut=PSONetwork(NewW1,NewB1,NewW2,NewB2,TestSamIn);
        AllTestNNOut=[AllTestNNOut;TestNNOut];
    end
    for i=1:TrainSamNum
        AverageTrainOut(HiddenUnitNum,i)=sum(AllTrainNNOut(:,i))/MaxRunningTime;
    end

    for i=1:TestSamNum
        AverageTestOut(HiddenUnitNum,i)=sum(AllTestNNOut(:,i))/MaxRunningTime;
    end
end
fid = fopen('result.txt','a+');
for HiddenUnitNum=3:MaxHiddenLayerNode;
    fprintf(fid,'\n ----------------------------');
    fprintf(fid,'\n the hidden unit number is  %i',HiddenUnitNum);
    for i=1:TestSamNum
        fprintf(fid,'\n %12.8f  %12.8f',TestSamOut(i),AverageTestOut(HiddenUnitNum,i));
    end
end
fclose(fid);
%
function [NewW1,NewB1,NewW2,NewB2]=PSOTrain(SamIn,SamOut,HiddenUnitNum);
Maxgeneration=700;  
E0=0.0001;
Xmin=-10;
Xmax=10;
Vmin=-5;
Vmax=5;
M=100;
c1=2.7;
c2=1.3;
w=0.9;
%
[R,SamNum]=size(SamIn);
[S2,SamNum]=size(SamOut);
generation=1;
Done=0;