Immune system method for constraint handling

The immune system method is a constraints handling method in which a immune system simulation is performed in order to reduce the number of non feasible individuals in the population through the action of an antibody population. In this tutorial, we will learn how to solve a constrained problem with this technique.

Method

The immune system method implemented in PaGMO/PyGMO has a similar design to the one of the co-evolution method. It uses two populations, the first population is associated to the initial problem from which the constraints are removed. The second population emulates the immune system where antibodies are evolved to match a certain number of antigens. These antigens are selected to be the best individuals, in term of feasibility, of the first population. Once found, the best antibodies are fed back into the first population. In PaGMO/PyGMO, the matching process is done with a simple algorithm associated with a problem that reduces the distance between antibodies and antigens. This distance or matching function is either the Hamming or Euclidean distance. It means that the matching process does not require additional evaluations of the objective or constraints functions. This step is thus computationaly efficient. The final implementation is based on a meta-algorithm, that takes the initial population to be optimized and two algorithms to evolve both the population associated to the modified problem and the immune system. In the following we are going to see how to use this constraints handling technique.

Application

The problem considered here is the problem g06 from the Congress on Evolutionary Computation 2006 (CEC2006). This problem has a cubic objective function with two non linear inequality constraints. To solve this problem the Differential Evolution (DE) algorithm is used for both the first and the second population. The number of iterations for the first algorithm must be set to 1 as the overall number of iterations is driven by the meta-algorithm itself. The number of iteration of the meta-algorithm is set to 5000, however the algorithm will stop as soon as it reaches convergence. The number of iterations for the immune system is set to 70 and a initial population of 90 individuals is chosen.

First import the PyGMO library and choose the population size and the number of generations for the meta-algorithm.

In [1]: from PyGMO import *
In [2]: pop_size = 90
In [3]: n_gen = 5000
In [4]: n_immune_gen = 70

Then creates the algorithms you wish to use for both populations. Here we have decided to use the Differential Evolution for both populations.

In [5]: algo_1 = algorithm.de(gen = 1, xtol=1e-30, ftol=1e-30)
In [6]: algo_2 = algorithm.de(gen = n_immune_gen, xtol=1e-30, ftol=1e-30)

Select the problem and associate a population to this problem.

In [7]: prob = problem.cec2006(6)
In [8]: pop = population(prob,pop_size)

Creates the meta-algorithm with these informations.

In [9]: algo_meta = algorithm.cstrs_immune_system(algorithm = algo_1, algorithm_immune = algo_2, gen = n_gen,select_method = algorithm.cstrs_immune_system.select_method.INFEASIBILITY, inject_method = algorithm.cstrs_immune_system.inject_method.CHAMPION, distance_method = algorithm.cstrs_immune_system.distance_method.EUCLIDEAN)

Here we have selected the infeasibility method where the antigen population is set by selecting individuals based on their infeasibility. The original selection, where only the best infeasible individual is selected for the population, from Coello did not give satisfactory results on this problem. The injected antibodies are a copy of the champion and the distance to evolve the antibodies is the Euclidean distance.

Evolve the population with the mat-algorithm described.

In [10]: pop = algo_meta.evolve(pop)

And finally, print the solutions.

In [11]: print(pop.champion.x)
In [12]: print(pop.champion.f)
In [13]: print(pop.champion.c)
Out [1]:
(14.094999999999994, 0.8429607892154646)
(-6961.813875580156,)
(-3.552713678800501e-15, 0.0)

As a comparison, the best known solution can be printed for this particular problem:

In [11]: print(prob.best_x)
In [12]: print(prob.best_f)
In [13]: print(prob.best_c)
Out [2]:
((14.095, 0.8429607892154796),)
((-6961.813875580138,),)
((-7.105427357601002e-15, 0.0),)

As seen, the algorithm has converged to the optimal constrained solution.