Genetic Algorithms For Multi-objective Optimization Problems

Many optimization problems in Scientific research and engineering practice can be modeled as multi-objective optimization problems. Thus, multi-objective optimization problems (MOP) have a wide range of applications, and designing effective algorithms for them is of not only the great importance in scientific research, but also the great value in applications.The main works in this thesis are as follows:Weighted genetic algorithms for multi objective optimization problem is attempt to aggregate all the objectives into one objective function with parameters. It has great importance to determine the weights in the multi-objective optimization evaluation index system. So the reliability and validity of the multi-objective optimization results are related to whether the attribute weights are determined scientifically and reasonably. First, focuses on the weighted sum genetic algorithm, the initial population are generated by a uniform design method, and after the objectives are normalized, the fitness function is defined by weighted sum of the normalized objectives, where the weights are dynamically defined. Based on these, a new multi-objective genetic algorithm called NWMOGA is proposed for multi-objective optimization problems.Second, Genetic algorithms can conduct several researches simultaneously, but the searches for the non-dominated solution are not always uniform. To search for more uniform distribution of non-dominated solutions, a new genetic algorithm for multi-objective optimization problems based on uniform design called BUMOGA is proposed combined with uniform design. The algorithm can find the sparse areas of non-dominated frontier, and explore the sparse area which can make the non-dominated solutions more uniform. The introductions of uniform crossover operator and single point crossover complex operator make up the defects of weak search capabilities of simulated binary crossover operator. The global convergence of the algorithm is proved, and effectiveness of the algorithm is demonstrated by the simulations.