1、%與傳統(tǒng)的學(xué)習(xí)算法不同，單隱層前饋神經(jīng)網(wǎng)絡(luò)(SLFNs) 極限學(xué)習(xí)機(jī) (ELM) 對(duì)輸入權(quán)值進(jìn)行動(dòng)態(tài)選擇%清空環(huán)境變量elmclcclearclose allformat compact rng( 'default' )%導(dǎo)入數(shù)據(jù)% file,path = uigetfile('*.xlsx','Select One or MoreFiles', 'MultiSelect', 'on');% filename=path file;filename= 'CPSO 優(yōu)化 ELM 實(shí)現(xiàn)分類(lèi) 最終版本訓(xùn)練集.xls

2、x'M=xlsread(filename);input=M(:,1:end-1);output=M(:,end);%數(shù)據(jù)預(yù)處理inputn,maps=mapminmax(input',0,1);% 輸出標(biāo)簽轉(zhuǎn)換為 one hot 標(biāo)簽outputn=one_hot(output);n_samples=size(inputn,2);n=randperm(n_samples);m=floor(0.7*n_samples);Pn_train=inputn(:,n(1:m);Tn_train=outputn(:,n(1:m);Pn_valid=inputn(:,n(m+1:end);T

3、n_valid=outputn(:,n(m+1:end);%節(jié)點(diǎn)個(gè)數(shù)inputnum=size(Pn_train,1);%輸入層節(jié)點(diǎn)hiddennum=160;%隱含層節(jié)點(diǎn)type= 'sig' ; %sin %hardlim %sig% 隱含層激活函數(shù)% ticIW,B,LW,TF =elmtrain(Pn_train,Tn_train,hiddennum,'sig' );TY2 = elmpredict(Pn_valid,IW,B,LW,TF); toc% 計(jì)算分類(lèi)概率prob0=prob_cal(TY2);%第一行為屬于0 (不滑坡)的概率第二行為屬于 1

4、 (滑坡)的概率，上下兩個(gè)概率和為 1% 看看準(zhǔn)確率% 驗(yàn)證集分類(lèi)結(jié)果,J=max(Tn_valid);,J1=max(prob0);disp( ' 優(yōu)化前 ' )accuracy=sum(J=J1)/length(J)TY2 = elmpredict(Pn_train,IW,B,LW,TF);% 計(jì)算分類(lèi)概率prob0=prob_cal(TY2);%第一行為屬于0 (不滑坡)的概率第二行為屬于 1 (滑坡)的概率，上下兩個(gè)概率和為 1% 看看準(zhǔn)確率% 驗(yàn)證集分類(lèi)結(jié)果,J=max(Tn_train);,J1=max(prob0);disp( ' 優(yōu)化前 ' )a

5、ccuracy=sum(J=J1)/length(J) function IW,B,LW,TF = elmtrain(P_train,T,N,TF)%P_train=train;T=train_label;N=200;TF='sig'% P - Input Matrix of Training Set (R*Q)% T - Output Matrix of Training Set (S*Q)% N - Number of Hidden Neurons (default = Q)% TF - Transfer Function:'sig' for Sigmoid

6、al function (default)',sin' for Sine function,'hardlim' for Hardlim function% IW - Input Weight Matrix (N*R)% B - Bias Matrix (N*1)% LW - Layer Weight Matrix (N*S)% IW,B,LW,TF = elmtrain(P,T,20,'sig')% Y = elmtrain(P,IW,B,LW,TF)% See also ELMPREDICT%輸出樣本數(shù)量必須與輸出樣本數(shù)量一致。R,Q = si

7、ze(P_train);%R=2,Q=1900S,Q = size(T);%S=1,Q=1900% 隨機(jī)產(chǎn)生輸入權(quán)重矩陣， 1900*2% rand('seed',sum(100*clock)IW = rand(N,R) * 2 - 1;% 隨機(jī)產(chǎn)生隱層偏置1900*1% rand('seed',sum(100*clock)B = rand(N,1);BiasMatrix = repmat(B,1,Q);% 計(jì)算隱層輸出 HtempH = IW * P_train + BiasMatrix;switch TFcase 'sig'H = 1 ./ (1 + exp(-tempH);case 'sin'H = sin(tempH);case 'hardlim'H = hardlim(tempH);end% 計(jì)算隱層到輸出層之間的權(quán)重LW = pinv(H') * T'endfunction y=one_hot(x)class=unique(x);分類(lèi) ' )disp( ' 此數(shù)據(jù)為 ' ,num2str(length(class),if min(x)=0x=x+1;elsex=x;e