遗传算法的适应性拒绝让步

Question

我正在制作一个遗传算法来将字符演变成“Hello World”，但是当我执行程序时，它会跨越一次并在剩余的运行中保持不变。我确定我的 crossOver 和 mutate 方法有效，但我的 chooseParents 方法有问题吗？

    public class Algorithm{
       private static int newPopLength = 5;
       private static final double MUTATE_RATE = .015;
       private static final double CROSSOVER_RATE = .6;

       public static Population evolvePopulation(Population pop){
          Population newPopulation = new Population(pop.size());
       //crossover
          for(int  x = 0; x<pop.size(); x++){
          //chooses two parents for crossover
             Chromosome chrom1 = chooseParents(pop);
             Chromosome chrom2 = chooseParents(pop);
             Chromosome newChild = crossOver(chrom1, chrom2);
             newPopulation.setChromosome(x, newChild);
          }
       //mutate
          for(int x = 0; x < pop.size(); x++){
             mutate(newPopulation.getChromosome(x));
          }
          return newPopulation;
       } 

       //crossover method
       public static Chromosome crossOver(Chromosome chrom1, Chromosome chrom2){
          Chromosome childChrom = new Chromosome();
          childChrom.generateChromosome();
       //Creates and generates childChromosome

          char []arr1 = chrom1.getChromosome();
          char []arr2 = chrom2.getChromosome();
          char []childArr = new char[arr1.length];

       //crosses over by 1/2 of each array
          for(int x = 0; x<arr2.length; x++){
             if(x <= Math.round(arr1.length / 2)){
                childArr[x] = arr1[x];
             }
             else{
                childArr[x] = arr2[x];
             }
          }
          for(int x = 0; x<childArr.length; x++){
             childChrom.setGene(childArr[x], x);
          }
          return childChrom;
       }
       //mutates chromosome by selecting a random point and replacing it with a random char
       public static void mutate(Chromosome chrom){
          if(Math.random() <= MUTATE_RATE){
             int rand = Math.round((int)(Math.random() * chrom.size()));
             chrom.setGene((char)((Math.random()*25) + 97), rand);
          }
       }
       private static Chromosome chooseParents(Population pop){
          Population newPopulation = new Population(newPopLength);

          Chromosome fittest = new Chromosome();
          for(int x = 0; x<newPopulation.size(); x++){
          //randomlu chooses 5 chromosomes
             Chromosome newChrom = pop.getChromosome((int)(Math.random()*pop.size()));
             newPopulation.setChromosome(x, newChrom);

          }
          return newPopulation.getFittest();

       }
    }
public class FitnessCalc{
   private static char solution[] = new char[Chromosome.getDefaultLength()];
   public static void setSolution(String word){
      solution = word.toCharArray();
   }

   public static char[] getSolution(){
      return solution;
   }
   public static int getFitness(Chromosome chrom){
      int fitness = 0;
      String chromWord = String.valueOf(chrom);
      char []chromArray = chromWord.toCharArray(); 

      for(int x = 0; x< solution.length;x++){
         //return 
fitness += Math.abs((int)(chrom.getGene(x)) - (int)(solution[x]));
      }
      return fitness;
   }


   public static int maxFitness(){
      int maxFitness =10241024;
      return maxFitness;
   }
}


   public Chromosome getFittest(){
      Chromosome fittest = new Chromosome();
      fittest.generateChromosome();
      for(int x = 0; x<size(); x++){
   if(FitnessCalc.getFitness(getChromosome(x))<= FitnessCalc.getFitness(fittest)){
         //if(FitnessCalc.getFitness(getChromosome(x)) >= FitnessCalc.getFitness(fittest)){
            fittest =  getChromosome(x);
         }
      }

      return fittest;
   }

输出：

Generation  992 Fittest Hello2aorld Fitness 28
Generation  993 Fittest Hello2aorld Fitness 28
Generation  994 Fittest Hello2aorld Fitness 28
Generation  995 Fittest Hello2aorld Fitness 28
Generation  996 Fittest Hello2aorld Fitness 28
Generation  997 Fittest Hello2aorld Fitness 28
Generation  998 Fittest Hello2aorld Fitness 28
Generation  999 Fittest Hello2aorld Fitness 28
Generation 1000 Fittest Hello2aorld Fitness 28


Generation 998 Fittest 2ello'aorld Fitness 39
Generation 999 Fittest 2ello'aorld Fitness 39
Generation 1000 Fittest 2ello'aorld Fitness 39

Generation 998 Fittest Sello,Porld Fitness 30
Generation 999 Fittest Sello,Porld Fitness 30
Generation 1000 Fittest Sello,Porld Fitness 30

Answer 1

让我们仔细看看你的静态getFitness(Chromosome chrom)函数的sn-p：

for(int x = 0; x< solution.length;x++){
  return fitness += Math.abs((int)(chrom.getGene(x)) - (int)(solution[x]));
}
return fitness;

你想在这里积累健身吗？如果是这样，它不起作用，因为您在添加绝对差异后立即返回健身，而 x = 0。您并没有遍历那里的所有基因，因此您可能应该摆脱第一个返回。

我感到困惑的另一件事是您的getFittest() 函数。这里你说最适合的染色体具有最高的适应度值：

if(FitnessCalc.getFitness(getChromosome(x)) >= FitnessCalc.getFitness(fittest)){
  fittest =  getChromosome(x);
}

但这真的是真的吗？请记住，您正在减去字符基因型（它们的 Int 表示），因此最适合的将是差异较小的那个，因为它更接近实际字符。根据经验，大多数优化问题都是关于找到最小值，因此您应该始终仔细检查。

尝试解决这两个问题，如果您在进化过程中仍然遇到问题，请告诉我们。