1 簡介

為了提高核極限學(xué)習(xí)機(jī)(ELM)的分類正確率,采用哈里斯鷹算法(HHO)對懲罰系數(shù),寬度參數(shù)兩個參數(shù)進(jìn)行優(yōu)化.首先,根據(jù)乳腺良惡性腫瘤數(shù)據(jù)庫訓(xùn)練集并利用哈里斯鷹算法優(yōu)化核極限學(xué)習(xí)機(jī);然后,通過HHO-ELM和ELM對測試集進(jìn)行分類診斷;最后,對比分析HHO-ELM和ELM的分類性能,測試結(jié)果表明,HHO-ELM的總體診斷正確率相較于ELM提高了10%,且惡性腫瘤的診斷正確率明顯優(yōu)于ELM.

2 部分代碼

function [fbst, xbst, performance] = hho( objective, d, lmt, n, T, S)%Harris hawks optimization algorithm% inputs: % ? objective - function handle, the objective function% ? d - scalar, dimension of the optimization problem% ? lmt - d-by-2 matrix, lower and upper constraints of the decision varable% ? n - scalar, swarm size% ? T - scalar, maximum iteration% ? S - scalar, times of independent runs% data: 2021-05-09% author: elkman, github.com/ElkmanY/%% Levy flightbeta = 1.5;sigma = ( gamma(1+beta)*sin(pi*beta/2)/gamma((1+beta)/2)*beta*2^((beta-1)/2) ).^(1/beta);Levy = @(x) 0.01*normrnd(0,1,d,x)*sigma./abs(normrnd(0,1,d,x)).^(1/beta);%% algorithm proceduretic;for s = 1:S ? ?%% ?Initialization ? ?X = lmt(:,1) + (lmt(:,2) - lmt(:,1)).*rand(d,n); ? ?for t = 1:T ? ? ? ?F = objective(X); ? ? ? ?[f_rabbit(s,t), i_rabbit] = min(F); ? ? ? ?x_rabbit(:,t,s) = X(:,i_rabbit); ? ? ? ?xr = x_rabbit(:,t,s); ? ? ? ?J = 2*(1-rand(d,1)); ? ? ? ?E0 = 2*rand(1,n)-1; ? ? ? ?E(t,:) = 2*E0*(1-t/T); ? ? ? ?absE = abs(E(t)); ? ? ? ?p1 = absE>=1; ? %eq(1) ? ? ? ?r = rand(1,n); ? ? ? ?p2 = (r>=0.5) & (absE>=0.5) & (absE<1); %eq(4) ? ? ? ?p3 = (r>=0.5) & (absE<0.5); ?%eq(6) ? ? ? ?p4 = (r<0.5) & (absE>=0.5) & (absE<1); %eq(10) ? ? ? ?p5 = (r<0.5) & (absE<0.5); %eq(11) ? ? ? ?%% update locations ? ? ? ?rh = randi([1,n],1,n); ? ? ? ?flag1 = rand(1,n)>=0.5; ? ? ? ?Y = xr - E(t,:).*abs( J.*xr - X ); ? ? ? ?Z = Y + rand(d,n).*Levy(n); ? ? ? ?flag2 = (objective(Y)<objective(Z)) & (objective(Y)<F); ? ? ? ?flag3 = (objective(Y)>objective(Z)) & (objective(Z)<F); ? ? ? ?flag4 = (~flag2) & (~flag3); ? ? ? ?X_ = ? ?p1.*( ? (X(:,rh) - rand(1,n).*abs( X(:,rh) - 2*rand(1,n).*X )).*flag1 +... ? ? ? ? ? ?((X(:,rh) - mean(X)) - rand(1,n).*( lmt(:,1) + (lmt(:,2) - lmt(:,1)).*rand(d,n) )).*(~flag1) ? )... ? ? ? ? ? ?+ ? p2.*( ? xr - X - E(t,:).*abs( J.*xr - X ) ? )... ? ? ? ? ? ?+ ? p3.*( ? xr - E(t,:).*abs( xr - X ) ? )... ? ? ? ? ? ?+ ? p4.*( ? Y.*flag2 + Z.*flag3 + ( lmt(:,1) + (lmt(:,2) - lmt(:,1)).*rand(d,n) ).*flag4 ?)... ? ? ? ? ? ?+ ? p5.*( ? Y.*flag2 + Z.*flag3 + ( lmt(:,1) + (lmt(:,2) - lmt(:,1)).*rand(d,n) ).*flag4 ?); ? ? ? ?X_(:,i_rabbit) = xr; ? ? ? ?X = X_; ? ?endend%% ê?3?-outputsperformance = [min(f_rabbit(:,T));mean(f_rabbit(:,T));std(f_rabbit(:,T))];timecost = toc;[fbst, ibst] = min(f_rabbit(:,T));xbst = x_rabbit(:,T,ibst);%% ??í?-plot data% Convergence Curvefigure('Name','Convergence Curve');box onsemilogy(1:T,mean(f_rabbit,1),'b','LineWidth',1.5);xlabel('Iteration','FontName','Aril');ylabel('Fitness/Score','FontName','Aril');title('Convergence Curve','FontName','Aril');if d == 2 ? ?% Trajectory of Global Optimal ? ?figure('Name','Trajectory of Global Optimal'); ? ?x1 = linspace(lmt(1,1),lmt(1,2)); ? ?x2 = linspace(lmt(2,1),lmt(2,2)); ? ?[X1,X2] = meshgrid(x1,x2); ? ?V = reshape(objective([X1(:),X2(:)]'),[size(X1,1),size(X1,1)]); ? ?contour(X1,X2,log10(V),100); % notice log10(V) ? ?hold on ? ?plot(x_rabbit(1,:,1),x_rabbit(2,:,1),'r-x','LineWidth',1); ? ?hold off ? ?xlabel('\it{x}_1','FontName','Time New Roman'); ? ?ylabel('\it{x}_2','FontName','Time New Roman'); ? ?title('Trajectory of Global Optimal','FontName','Aril');endend

3 仿真結(jié)果

4 參考文獻(xiàn)

[1]彭甜, 孫偉, 張楚,等. 基于改進(jìn)哈里斯鷹算法優(yōu)化ELM的風(fēng)速預(yù)測方法及系統(tǒng):.?

博主簡介：擅長智能優(yōu)化算法、神經(jīng)網(wǎng)絡(luò)預(yù)測、信號處理、元胞自動機(jī)、圖像處理、路徑規(guī)劃、無人機(jī)等多種領(lǐng)域的Matlab仿真，相關(guān)matlab代碼問題可私信交流。

部分理論引用網(wǎng)絡(luò)文獻(xiàn)，若有侵權(quán)聯(lián)系博主刪除。