100基于改进ISODATA算法的负荷场景曲线聚类matlab.rar

function [mean1,class1]=ISODATA_1(x,K,theta_N,theta_S,theta_c,L) %% step1 xx=x;I2=2; mean1= cell(K,1); mean1{1}=x(1,:); for i=2:K ds(i)=Distance1(xx(i,:),mean1{1}); [v1,v2]=max(ds); mean1{i}=xx(v2,:); xx(v2,:)=[]; end n = size(x,1); N_c = K; ite = 1; while ite<I2 flag = 1; while flag %% step2 class1 = cell(size(mean1));MM=0; for i=1:n num = Belong1(x(i,:),mean1); class1{num} = [class1{num};x(i,:)]; end for i=1:6 if isempty(class1{i}) MM=MM+1; end end %% step3 for i=1:N_c size_i = size(class1{i},1); if size_i<theta_N class_i = class1{i}; mean1 = DeleteRow(mean1,i); class1 = DeleteRow(class1,i); N_c = N_c-1; for j=1:size_i class_ij = class_i(j,:);%the j'th row of class{i} num = Belong1(class_ij,mean1); class1{num} = [class1{num};class_ij]; end end end %% step4 for i=1:N_c if ~isempty(mean1{i}) mean1{i} = sum(class1{i})./size(class1{i},1); end end %% step5 Dis = zeros(N_c,1); for i=1:N_c if ~isempty(class1{i}) N_i =size(class1{i},1); tmp = bsxfun(@minus,class1{i},mean1{i}); Dis(i) = sum(arrayfun(@(x)norm(tmp(x,:)),1:N_i))/N_i; end end %% step6 D = 0; for i=1:N_c if ~isempty(class1{i}) N_i =size(class1{i},1); D = D + N_i*Dis(i); end end D = D/n; %% step7 flag = 0; if ite == I2 theta_c = 0; flag = 0; elseif ~(N_c > K/2) flag = 1; elseif mod(ite,2)==0 || ~(N_c<2*K) flag = 0; end %% 分裂处理 %% step8 if flag flag = 0; delta = cell(N_c,1); for i=1:N_c if ~isempty(class1{i}) N_i =size(class1{i},1); tmp = bsxfun(@minus,class1{i},mean1{i}); delta{i} = arrayfun(@(x)norm(tmp(:,x)),1:size(tmp,2))/N_i; end end %% step9 delta_max = cell(N_c,1); for i=1:N_c if ~isempty(class1{i}) max_i = max(delta{i}); sub = find(delta{i}==max_i,1); delta_max{i} = [max_i,sub]; end end %% step10 for i=1:N_c if delta_max{i}(1) > theta_S N_i =size(class1{i},1); con1 = (Dis(i)>D && N_i>2*(theta_N + 1)); con2 = ~(N_c>K/2); if con1 || con2 %%%%这里分裂%%%%% flag = 1;%一旦发生分裂，那么分裂一次后就返回第二步；若没发生分裂，则直接进入合并处理步 lamda = 0.5; max_sub = delta_max{i}(2); mean1{i}(max_sub) = mean1{i}(max_sub) + lamda * delta_max{i}(1); addOneMean = mean1{i}; addOneMean(max_sub) = addOneMean(max_sub) - lamda * delta_max{i}(1); mean1 = [mean1;addOneMean]; N_c = N_c+1; break; end end end end end %% 合并处理 if L %% step11 Distance = zeros(N_c,N_c); for i=1:N_c-1 for j=i:N_c Distance(i,j) = norm(mean1{i}-mean1{j}); end end %% step12 index = find(-Distance>theta_c); keepIndex = [Distance(index),index]; [~, index] = sort(keepIndex(:,1)); if size(index,1) > L index = index(1:L,:); end %% step13 if size(index,1) ~= 0 for id=1:size(index,1) [m_i m_j]= seq2idx(index(id),N_c); %%%%%这里合并%%%%% N_mi = size(class1{m_i},1); N_mj = size(class1{m_j},1); mean1{m_i} = (N_mi*mean1{m_i} + N_mj*mean1{m_j})/(N_mi+N_mj); mean1 = DeleteRow(mean1,m_j); class1{m_i} = [class1{m_i};class1{m_j}]; class1 = DeleteRow(class1,m_j); end end end %% step14 ite=ite+1; end end