1 tic % 計時器 2 %% 清空環境變量 3 close all 4 clear 5 clc 6 format compact 7 %% 數據提取 8 % 載入測試數據wine,其中包含的數據為classnumber = 3,wine:178*13的矩陣,wine_labes:178*1的列向量 9 load wine.mat 10 % 選定訓練集和測試集 11 % 將第一類的1-30,第二類的60-95,第三類的131-153做為訓練集 12 train_wine = [wine(1:30,:);wine(60:95,:);wine(131:153,:)]; 13 % 相應的訓練集的標簽也要分離出來 14 train_wine_labels = [wine_labels(1:30);wine_labels(60:95);wine_labels(131:153)]; 15 % 將第一類的31-59,第二類的96-130,第三類的154-178做為測試集 16 test_wine = [wine(31:59,:);wine(96:130,:);wine(154:178,:)]; 17 % 相應的測試集的標簽也要分離出來 18 test_wine_labels = [wine_labels(31:59);wine_labels(96:130);wine_labels(154:178)]; 19 %% 數據預處理 20 % 數據預處理,將訓練集和測試集歸一化到[0,1]區間 21 [mtrain,ntrain] = size(train_wine); 22 [mtest,ntest] = size(test_wine); 23
24 dataset = [train_wine;test_wine]; 25 % mapminmax為MATLAB自帶的歸一化函數 26 [dataset_scale,ps] = mapminmax(dataset',0,1);
27 dataset_scale = dataset_scale';
28
29 train_wine = dataset_scale(1:mtrain,:); 30 test_wine = dataset_scale( (mtrain+1):(mtrain+mtest),: ); 31 %% 利用灰狼算法選擇最佳的SVM參數c和g 32 SearchAgents_no=10; % 狼群數量，Number of search agents 33 Max_iteration=10; % 最大迭代次數，Maximum numbef of iterations 34 dim=2; % 此例需要優化兩個參數c和g，number of your variables 35 lb=[0.01,0.01]; % 參數取值下界 36 ub=[100,100]; % 參數取值上界 37 % v = 5; % SVM Cross Validation參數,默認為5 38
39 % initialize alpha, beta, and delta_pos 40 Alpha_pos=zeros(1,dim); % 初始化Alpha狼的位置 41 Alpha_score=inf; % 初始化Alpha狼的目標函數值，change this to -inf for maximization problems 42
43 Beta_pos=zeros(1,dim); % 初始化Beta狼的位置 44 Beta_score=inf; % 初始化Beta狼的目標函數值，change this to -inf for maximization problems 45
46 Delta_pos=zeros(1,dim); % 初始化Delta狼的位置 47 Delta_score=inf; % 初始化Delta狼的目標函數值，change this to -inf for maximization problems 48
49 %Initialize the positions of search agents 50 Positions=initialization(SearchAgents_no,dim,ub,lb); 51
52 Convergence_curve=zeros(1,Max_iteration); 53
54 l=0; % Loop counter循環計數器 55
56 % Main loop主循環 57 while l<Max_iteration % 對迭代次數循環 58 for i=1:size(Positions,1) % 遍歷每個狼 59
60 % Return back the search agents that go beyond the boundaries of the search space 61 % 若搜索位置超過了搜索空間，需要重新回到搜索空間 62 Flag4ub=Positions(i,:)>ub; 63 Flag4lb=Positions(i,:)<lb; 64 % 若狼的位置在最大值和最小值之間，則位置不需要調整，若超出最大值，最回到最大值邊界； 65 % 若超出最小值，最回答最小值邊界 66 Positions(i,:)=(Positions(i,:).*(~(Flag4ub+Flag4lb)))+ub.*Flag4ub+lb.*Flag4lb; % ~表示取反 67
68 % 計算適應度函數值 69 cmd = [' -c ',num2str(Positions(i,1)),' -g ',num2str(Positions(i,2))]; 70 model=svmtrain(train_wine_labels,train_wine,cmd); % SVM模型訓練 71 [~,fitness]=svmpredict(test_wine_labels,test_wine,model); % SVM模型預測及其精度 72 fitness=100-fitness(1); % 以錯誤率最小化為目標 73
74 % Update Alpha, Beta, and Delta 75 if fitness<Alpha_score % 如果目標函數值小於Alpha狼的目標函數值 76 Alpha_score=fitness; % 則將Alpha狼的目標函數值更新為最優目標函數值，Update alpha 77 Alpha_pos=Positions(i,:); % 同時將Alpha狼的位置更新為最優位置 78 end 79
80 if fitness>Alpha_score && fitness<Beta_score % 如果目標函數值介於於Alpha狼和Beta狼的目標函數值之間 81 Beta_score=fitness; % 則將Beta狼的目標函數值更新為最優目標函數值，Update beta 82 Beta_pos=Positions(i,:); % 同時更新Beta狼的位置 83 end 84
85 if fitness>Alpha_score && fitness>Beta_score && fitness<Delta_score % 如果目標函數值介於於Beta狼和Delta狼的目標函數值之間 86 Delta_score=fitness; % 則將Delta狼的目標函數值更新為最優目標函數值，Update delta 87 Delta_pos=Positions(i,:); % 同時更新Delta狼的位置 88 end 89 end 90
91 a=2-l*((2)/Max_iteration); % 對每一次迭代，計算相應的a值，a decreases linearly fron 2 to 0
92
93 % Update the Position of search agents including omegas 94 for i=1:size(Positions,1) % 遍歷每個狼 95 for j=1:size(Positions,2) % 遍歷每個維度 96
97 % 包圍獵物，位置更新 98
99 r1=rand(); % r1 is a random number in [0,1] 100 r2=rand(); % r2 is a random number in [0,1] 101
102 A1=2*a*r1-a; % 計算系數A，Equation (3.3) 103 C1=2*r2; % 計算系數C，Equation (3.4) 104
105 % Alpha狼位置更新 106 D_alpha=abs(C1*Alpha_pos(j)-Positions(i,j)); % Equation (3.5)-part 1
107 X1=Alpha_pos(j)-A1*D_alpha; % Equation (3.6)-part 1
108
109 r1=rand(); 110 r2=rand(); 111
112 A2=2*a*r1-a; % 計算系數A，Equation (3.3) 113 C2=2*r2; % 計算系數C，Equation (3.4) 114
115 % Beta狼位置更新 116 D_beta=abs(C2*Beta_pos(j)-Positions(i,j)); % Equation (3.5)-part 2
117 X2=Beta_pos(j)-A2*D_beta; % Equation (3.6)-part 2
118
119 r1=rand(); 120 r2=rand(); 121
122 A3=2*a*r1-a; % 計算系數A，Equation (3.3) 123 C3=2*r2; % 計算系數C，Equation (3.4) 124
125 % Delta狼位置更新 126 D_delta=abs(C3*Delta_pos(j)-Positions(i,j)); % Equation (3.5)-part 3
127 X3=Delta_pos(j)-A3*D_delta; % Equation (3.5)-part 3
128
129 % 位置更新 130 Positions(i,j)=(X1+X2+X3)/3;% Equation (3.7) 131
132 end 133 end 134 l=l+1; 135 Convergence_curve(l)=Alpha_score; 136 end 137 bestc=Alpha_pos(1,1); 138 bestg=Alpha_pos(1,2); 139 bestGWOaccuarcy=Alpha_score; 140 %% 打印參數選擇結果 141 disp('打印選擇結果'); 142 str=sprintf('Best Cross Validation Accuracy = %g%%，Best c = %g，Best g = %g',bestGWOaccuarcy*100,bestc,bestg); 143 disp(str) 144 %% 利用最佳的參數進行SVM網絡訓練 145 cmd_gwosvm = ['-c ',num2str(bestc),' -g ',num2str(bestg)]; 146 model_gwosvm = svmtrain(train_wine_labels,train_wine,cmd_gwosvm); 147 %% SVM網絡預測 148 [predict_label,accuracy] = svmpredict(test_wine_labels,test_wine,model_gwosvm); 149 % 打印測試集分類准確率 150 total = length(test_wine_labels); 151 right = sum(predict_label == test_wine_labels); 152 disp('打印測試集分類准確率'); 153 str = sprintf( 'Accuracy = %g%% (%d/%d)',accuracy(1),right,total); 154 disp(str); 155 %% 結果分析 156 % 測試集的實際分類和預測分類圖 157 figure; 158 hold on; 159 plot(test_wine_labels,'o'); 160 plot(predict_label,'r*'); 161 xlabel('測試集樣本','FontSize',12); 162 ylabel('類別標簽','FontSize',12); 163 legend('實際測試集分類','預測測試集分類'); 164 title('測試集的實際分類和預測分類圖','FontSize',12); 165 grid on 166 snapnow 167 %% 顯示程序運行時間 168 toc
