Application Research Of BP Neural Network Based On Genetic Algorithm In Multi-objective Optimization

Multi-objective optimization orientates from the design, modeling and programming of complicate systems including industrial manufacture, budget, network communication and road layout. In our daily life, a variety of constraints should be considered and the confliction among different objectives needs to be solved in almost every important decision and prediction. So this refers to the issue of multi-objective optimization. As these objectives are not independent, the optimization of one objective is achieved at the cost of the other objectives. The objectives have no common units and thus it is hard to evaluate the solution of multi-objective problems. Traditional methods have many disadvantages, for instance, the distribution of each objective's power is often quite subjective, the final objective is only represented by the weighted sum, and the optimization process can not be controlled.In this paper, we focus on the two aspects of artificial intelligence, genetic algorithm and artificial neural network. After the analysis, we try to combine these two technologies and apply it into the solving process of multi-objective optimization. The main contents of this thesis are as follows:Introduction of genetic algorithm theory. We will introduce the short- comings of current multi-objective genetic algorithm, and discuss major issues waited to be overcome, such as the allocation of fitness, diversity maintenance and convergence. Due to that simple genetic algorithm is hard to converge in practice, and often encounters a local optimum, so we bring forward an improved intercurrent mixed nondominated sorting genetic algorithm II method based on nondominated sorting genetic algorithm II. Several algorithm cases indicate that the improved method is obviously superior to nondominated sorting genetic algorithm II, not only in diversity of solution but also convergence.By analyzing the shortcomings of BP neural network, the study on key technologies and algorithm of BP neural network is conducted, and an improved method is proposed. Combining improved genetic algorithm and BP neural network, we put forward the BP neural network based on genetic algorithm. Taking advantage of both genetic algorithm's global search capability and BP algorithm's local optimize performance, improved method's convergence accuracy and rate are raised in solving multi-objective optimization. In practice, a favorable performance is achieved in the of multi-objective workshop dynamic scheduling.