This paper demonstrates that running a genetic algorithm with injection of individuals from solutions to similar problems can get better performance than running a genetic algorithm with random initialization. We also find that running GAs with information from the original problem or the adding one city problem can get better performance than running GAs with information from other kinds of modified problems. We believe this is because the same problem and the add1 problem contain all cities of the original problem and do not change the length of the edges and are therefore more similar to the original problems.
However, injecting individuals from the diff2 problem can sometimes help GAs get better performance than when injecting individuals from the diff1 problem. This may be because after changing a certain city or changing the location of a city over a certain percentage, the new problem is less similar to the original problem. In the future, we want to find out how can we define the modified problem and the original problem as similar. We want to know