教与学算法（Teaching Learning Based Optimization，TLBO）

Halcom

教与学算法（Teaching Learning Based Optimization，TLBO）

1. function [X, FVAL, BestFVALIter, pop] = ModifiedTLBO(FITNESSFCN,lb,ub,T,NPop)
   
2. % Teaching Learning Based optimization (TLBO)
   
3. % ModifiedTLBO attempts to solve problems of the following forms:
   
4. %         min F(X)  subject to: lb <= X <= ub
   
5. %          X        
   
6. %                           
   
7. %  [X,FVAL,BestFVALIter, pop] = ModifiedTLBO(FITNESSFCN,lb,ub,T,NPop)
   
8. %  FITNESSFCN   - function handle of the fitness function
   
9. %  lb           - lower bounds on X
   
10. %  ub           - upper bounds on X
    
11. %  T            - number of iterations
    
12. %  NPop         - size of the population (class size)  
    
13. %  X            - minimum of the fitness function determined by ModifiedTLBO
    
14. %  FVAL         - value of the fitness function at the minima (X)
    
15. %  BestFVALIter - the best fintess function value in each iteration
    
16. %  pop          - the population at the end of the specified number of iterations
    
17. 
    
18. % preallocation to store the best objective function of every iteration
    
19. % and the objective function value of every student
    
20. BestFVALIter = NaN(T,1);
    
21. obj = NaN(NPop,1);
    
22. 
    
23. % Determining the size of the problem
    
24. D = length(lb);
    
25. 
    
26. % Generation of initial population
    
27. pop = repmat(lb, NPop, 1) + repmat((ub-lb),NPop,1).*rand(NPop,D);
    
28. 
    
29. 
    
30. %  Evaluation of objective function
    
31. %  Can be vectorized
    
32. for p = 1:NPop
    
33.     obj(p) = FITNESSFCN(pop(p,:));
    
34. end
    
35. 
    
36. 
    
37. for gen = 1: T
    
38.    
    
39.     % Partner selection for all students
    
40.     % Note that randperm has been used to speedup the partner selection.
    
41.     Partner = randperm(NPop);
    
42.     % There is a remote possibility that the ith student will have itself as its partner
    
43.     % No experiment is available in literature on the disadvantages of
    
44.     % a solution having itself as partner solution.
    
45.    
    
46.     for i = 1:NPop
    
47.         
    
48.         % ----------------Begining of the Teacher Phase for ith student-------------- %
    
49.         mean_stud = mean(pop);
    
50.         
    
51.         % Determination of teacher
    
52.         [~,ind] = min(obj);
    
53.         best_stud = pop(ind,:);
    
54.         
    
55.         % Determination of the teaching factor
    
56.         TF = randi([1 2],1,1);
    
57.         
    
58.         % Generation of a new solution
    
59.         NewSol = pop(i,:) + rand(1,D).*(best_stud - TF*mean_stud);
    
60.         
    
61.         % Bounding of the solution
    
62.         NewSol = max(min(ub, NewSol),lb);
    
63.         
    
64.         % Evaluation of objective function
    
65.         NewSolObj = FITNESSFCN(NewSol);
    
66.         
    
67.         % Greedy selection
    
68.         if (NewSolObj < obj(i))
    
69.             pop(i,:) = NewSol;
    
70.             obj(i) = NewSolObj;
    
71.         end
    
72.         % ----------------Ending of the Teacher Phase for ith student-------------- %
    
73.         
    
74.         
    
75.         % ----------------Begining of the Learner Phase for ith student-------------- %
    
76.         % Generation of a new solution
    
77.         if (obj(i)< obj(Partner(i)))
    
78.             NewSol = pop(i,:) + rand(1, D).*(pop(i,:)- pop(Partner(i),:));
    
79.         else
    
80.             NewSol = pop(i,:) + rand(1, D).*(pop(Partner(i),:)- pop(i,:));
    
81.         end
    
82.         
    
83.         % Bounding of the solution
    
84.         NewSol = max(min(ub, NewSol),lb);
    
85.         
    
86.         % Evaluation of objective function
    
87.         NewSolObj =  FITNESSFCN(NewSol);
    
88.         
    
89.         % Greedy selection
    
90.         if(NewSolObj< obj(i))
    
91.             pop(i,:) = NewSol;
    
92.             obj(i) = NewSolObj;
    
93.         end
    
94.         % ----------------Ending of the Learner Phase for ith student-------------- %
    
95.         
    
96.     end
    
97.    
    
98.     % This is not part of the algorithm but is used to keep track of the
    
99.     % best solution determined till the current iteration
    
100.     [BestFVALIter(gen),ind] = min(obj);
     
101. end
     
102. 
     
103. % Extracting the best solution
     
104. X = pop(ind,:);
     
105. FVAL = BestFVALIter(gen);

复制代码

适应度函数：

1. function F = Rastrigin(X)
   
2. 
   
3. [ros, ~] = size(X);
   
4. F = zeros(ros, 1);
   
5. for k = 1: ros
   
6.     x = X(k,:);
   
7.     F(k,1) = sum((x.^2 - 10.*cos(2.*pi.*x) + 10));
   
8. end

复制代码

主函数：

1. rng(2,'twister')
   
2. 
   
3. FITNESSFCN = @Rastrigin;
   
4. 
   
5. lb = -5.12*ones(1,2);
   
6. ub = 5.12*ones(1,2);
   
7. 
   
8. NPop = 50;
   
9. T = 90;
   
10. 
    
11. [X,FVAL,BestFVALIter] = ModifiedTLBO(FITNESSFCN,lb,ub,T,NPop);
    
12. 
    
13. display(['The minimum point is ', num2str(X)])
    
14. display(['The fitness function value at the mimimum point is ', num2str(FVAL)])
    
15. D = length(lb);
    
16. display(['The number of fitness function evaluation is ', num2str(NPop+2*NPop*T)])
    
17. 
    
18. plot(1:T,BestFVALIter,'r*')
    
19. xlabel('Iteration Number')
    
20. ylabel('Value of Fitness function')
    
21. grid on

复制代码