基于Matlab实现粒子群算法SVM（源码+数据）.rar

clc,clear,close all % 清屏+清理工作区+关闭图形窗口 warning off % 取消警告 feature jit off % 加速通道 tic global popmin1 popmax1 popmin2 popmax2 nvar % 加载数据 load('data.mat') data0 = data; for i=1:size(data,2) % 样本归一化 minmax(i,1) = min(data(:,i)); minmax(i,2) = max(data(:,i)); data(:,i) = (minmax(i,2)-data(:,i))./(minmax(i,2)-minmax(i,1)); end %% 构造训练预测数据 train_data = data(136:end,1:4); train_output_data = data(136:end,5); test_data = data(1:135,1:4); test_output_data = data(1:135,5); %% 粒子群PSO算法参数设置 maxgen = 20; % 最大迭代次数 sizepop = 5; % 种群数量 Vmax = 1; % 粒子速度上限 Vmin = -1; % 粒子速度下限 c1 = 1.4995; % 学习因子1 c2 = 1.4995; % 学习因子2 % 变量坐标取值范围 nvar = 2; % 未知量数量 popmin1 = 0.1; popmax1 = 2; % SVM惩罚系数C popmin2 = 0.1; popmax2 = 2; % SVM径向基函数基宽sigma %% 初始化粒子群PSO种群位置 pop=[]; for i=1:sizepop pop1.C = unifrnd(popmin1,popmax1,1,1); % 均匀分布解 pop1.sigma = unifrnd(popmin2,popmax2,1,1); % 均匀分布解 pop = [pop; pop1]; fitness(i) = fun( pop(i,:),train_data,train_output_data ); end clear pop1 V = Vmax*rands(sizepop,2); % 初始化速度 [bestfitness,bR]= max(fitness); % 亮度最高的保留 zbest = pop(bR,:); % 全局最佳 fitnesszbest = bestfitness; % 全局最佳适应度值 gbest = pop; % 个体最佳 fitnessgbest = fitness; % 个体最佳适应度值 trace = zbest; % 记录最优的种群 %% 粒子群PSO算法迭代寻优 for i=1:maxgen disp(['Iteration ' num2str(i)]); % 计算适应度值 for j=1:sizepop % 速度更新 V(j,:) = V(j,:) + c1*rand*([gbest(j).C,gbest(j).sigma] - [pop(j).C,pop(j).sigma] ) +... c2*rand*([zbest.C,zbest.sigma] - [pop(j).C,pop(j).sigma] ); V(j,find(V(j,:)>Vmax))=Vmax; % 上限 V(j,find(V(j,:)<Vmin))=Vmin; % 下限 % 种群更新 pop(j).C = pop(j).C + 0.5*V(j,1); pop(j).sigma = pop(j).sigma + 0.5*V(j,2); % pop个体取值范围约束 if pop(j).C > popmax1 % 上限 pop(j).C = popmax1 ; elseif pop(j).C < popmin1 % 下限 pop(j).C = popmin1; end if pop(j).sigma > popmax2 % 上限 pop(j).sigma = popmax2 ; elseif pop(j).sigma < popmin2 % 下限 pop(j).sigma = popmin2; end fitness(j) = fun( pop(j,:) ,train_data,train_output_data ); % 计算适应度值 % 个体最优更新 if fitness(j) > fitnessgbest(j) gbest(j,:) = pop(j,:); fitnessgbest(j) = fitness(j); end % 群体最优更新 if fitness(j) > fitnesszbest zbest = pop(j,:); fitnesszbest = fitness(j); end end trace = [trace,zbest]; % 记录最优的种群 fitness_iter(i) = fitnesszbest; % 最优适应度值 end time = toc; disp(['CPU计算时间 = ' num2str(time)]) save PSO_iter_result.mat fitness_iter trace zbest gbest train_data train_output_data data0 data test_data test_output_data minmax figure(1), plot(fitness_iter,'b.-','linewidth',2);grid on; xlabel('迭代次数');ylabel('适应度值');axis tight; % 计算适应度值 [fitnessZ,predict_result]= fun_predict( zbest,train_data,train_output_data,test_data ); disp(['最优惩罚因子C = ' num2str(zbest.C)]) disp(['最优径向基函数基宽sigma = ' num2str(zbest.sigma)]) figure(2), plot(data0(136:end,end),'r.-','linewidth',2);axis tight; hold on % 反归一化 plot(minmax(end,2) - predict_result{1}*(minmax(end,2)-minmax(end,1)),'b.-','linewidth',2);axis tight; legend('原始训练样本','训练样本预测值');hold off; xlabel('样本序列');ylabel('预测值');axis tight; figure(3), plot(data0(1:135,end),'r.-','linewidth',2);axis tight; hold on % 反归一化 plot(minmax(end,2) - predict_result{2}*(minmax(end,2)-minmax(end,1)),'b.-','linewidth',2);axis tight; legend('原始测试样本','测试样本预测值');hold off; xlabel('样本序列');ylabel('预测值');axis tight; RMSE = sqrt(mean((data(1:135,end)'-predict_result{2}).^2)); % R2 = 1-sum(data(1:135,end)'-predict_result{2})./sum(data(1:135,end)-mean(data(1:135,end))); disp(['RMSE = ' num2str(RMSE)]) % disp(['R2 = ' num2str(R2)]) %% rmse迭代画图 for i=1:length(trace) gbest_iter = trace(:,i); [fitnessZ,predict_result]= fun_predict( gbest_iter,train_data,train_output_data,test_data ); RMSEiter = sqrt(mean((data(1:135,end)'-predict_result{2}).^2)); RMSE_iter(i) = RMSEiter; end figure(4), plot(RMSE_iter,'b.-','linewidth',2);grid on; xlabel('迭代次数');ylabel('RMSE值');axis tight;