Genetic algorithm solves TSP problem

Pseudo code for the core

population_cur = init_population()
# Calculate the fitness value of the current population
fitness = get_fitness(population_cur)
 while does not meet the termination condition:
	 # Keep part of the elite parent
	population_next = select_sorted_population(fitness, population_cur, elite_size)
	 # Hybrid
    for i in range(population_size):
    	     # Here you can add the hybridization probability yourself
                 p1, p2 = selection(fitness, 2) # Use roulette selection operator to randomly select two as parents
        child1, child2 = crossover(population_cur[p1], population_cur[p2])

                 # Mutate children
        if random.random() < p_mutation:
            child1 = mutations.select_best_mutaion(child1, distmat)
        if random.random() < p_mutation:
            child2 = mutations.select_best_mutaion(child2, distmat)
		
        population_next.append(child1)
        population_next.append(child2)
                 # At this time, there are a total population size of five quarters of individuals, from which the next generation population is selected
        population_next = select_sorted_population(get_fitness(population_next), population_next, population_size)
	 	 # Replacement
        population_cur = population_next

Use a better variant

For an individual A to be mutated, separatelySliding variation，Flip mutation，irgibnnm variation, The one with the best three variants¹

Sliding variation

Randomly generate two subscripts a, b (a<b); move s[a] behind s[b]

Demo animation:

Flip mutation

Generate two subscripts a, b at random; reverse the sequence between s[a] and s[b]

Flip mutation is also called 2 transform method, and demonstrates the animation:

irgibnnm variation

First perform a flip mutation on the sequence; randomly select a city and exchange it to the nearest city to the map distance, the neighborhood in the literature is ±5

The experimental results of this mutation method do not seem to be as good as the results shown in the literature, may I have implemented it wrong? Demo animation:

Operation and results

Just run the gsp_ga.py file

The data set uses 78 points², About 10s running on my computer, the optimal distance in the impression is about 5400m

The complete code of genetic algorithm

github complete code

reference

---

1. Improving TSP Solutions Using GA with a New Hybrid Mutation Based on Knowledge and Randomness ︎

2. Github simulated annealing solves tsp problem ︎