Research Of Adaptive Multi-objective Genetic Algorithm With Agent

In real life many optimization problems composed by multiple targets, and theseobjectives are usually contradictory and even conflicting, solutions of the problem hasan infinite number. The main goal of solving multi-objective optimization problem isto get a group of comprehensive and effective solutions, decision makers choose fromthe solutions to solve problems in order to achieve the most optimal results.In general, the traditional mathematical programming methods are dependent onthe mathematical characteristics of optimization function, the capability to searchtarget space is low, and the calculation is very large. Because of the genetic algorithmhas better robustness, nothing with the problem, and fast features, there is a bigadvantage in solving multi-objective optimization problems. Genetic algorithmsperform well in the smart computing, this article will combine it with agenttechnology, proposed improved algorithm to solve multi-objective optimizationproblem.This paper describes the basic concepts, the development process and theoreticalknowledge of genetic algorithm and multi-objective optimization problem, anddescribed several classic multi-objective genetic algorithms. Then based on the ideaof multi-agent proposed a new multi-agent evolutionary algorithm, On the selection ofcrossover operator, in order to increase the diversity of computing results we shouldchoose different crossover operator, which according to participate in the crossoveroperation of the relationship between the two individual fitness and average fitness. Inorder to make the algorithm have strong global search ability in the early evolution,we should use uniform distribution mutation operator. In order to make the algorithmhave strong local search ability in the late evolution, we should use gaussian mutationoperator. By testing function to test and analysis the algorithm performance,simulation graphics show that the proposed algorithm can find a larger number anddistribution of Pareto solutions, the algorithm performance is good. Finally putforward another agent multi-objective evolutionary algorithm is applied to networkshortest path problem, an improved roulette wheel selection method, makes the choiceof agent is more competitive. According to the relationship between the averagefitness of the population and the average fitness of the two individuals of thecrossover, we should choose different crossover operator in the crossover operation.which makes a better agent, experimental cases show that this method can get good performance.