Font Size: a A A

Research On Multi-objective Constrained Optimization Algorithm Based On Particle Swarm

Posted on:2016-05-30Degree:MasterType:Thesis
Country:ChinaCandidate:X L DingFull Text:PDF
GTID:2298330467977378Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Multi-objective optimization problem widely exists in scientific research and engineering practice. In this dissertation, particle swarm optimization algorithm is used to handle two types of problems in the field of multi-objective optimization, namely continuous constrained optimization problem and discrete constrained optimization problem.When to handle the first category, against the problems that traditional multi-objective particle swarm optimization algorithms do not fully take into account the density information of non-inferior solutions and the balance between global search ability and local search capability when to select optimal solutions and crop non-inferior solutions, convergence and diversity of solutions are poor, and the high complexity of the algorithm, this dissertation proposes a multi-objective particle swarm optimization algorithm based on adaptive network and dynamic crowding distance. The algorithm evenly divides the objective function space to the same spacing grid, statistics the numbers of particles to estimate the density of particles in each grid and designs fitness function through the density of particles to select optimal Pareto solutions. When to crop the solutions, variance information of particles is introduced and the strategy based on dynamic crowding distance is designed, in order to avoid the problem that eliminating all individuals of small crowding distance one time will deteriorate distribution of solutions. When to handle constraints, the concept of violate constraints degree is defined and feasible degree guidelines are proposed. Then, parameters are studied from the angle of theoretical analysis and experimental verification. Function test and algorithm application are both verify the effectiveness and superiority of multi-objective particle swarm optimization proposed.When to handle the second category, classic multi-objective flow shop scheduling problem as a research object, two objective functions which are the cost and the output value are optimized. Against problems that traditional multi-objective particle swarm optimization easily converge to the local minima, and uniformity of the solutions is not strong, this dissertation presents a multi-objective particle swarm algorithm based on Baldwinian learning strategy for improving local search capability of algorithm. Then, learning from the thought that crowding distance is able to improve of the diversity of solutions in NSGA-II algorithm, a strategy of dynamic crowding distance is designed to achieve the balance of local optimization and global optimization. When to construct the fitness function, in order to avoid the prblem that fixed weights may cause the confusion of pros and cons of the order among Pareto solutions, a fitness function of dynamic weight relative weighting method is designed. In the parametric study part, orthogonal design method is used to screen out appropriate parameters with relatively little time in the case of guarantee the quality of parameters. In the part of function experiment, the superiority of the improved algorithm is verifid by optimal percentage error and mean percentage error, and the effectiveness of the algorithm is verifid by the index based on the distance to the reference set. Finally, a practical multi-objective flow shop scheduling problem is uese to verify the feasibility of the algorithm.
Keywords/Search Tags:multi-objective particle swarm, adaptive network, dynamic crowdingdistance, Baldwinian learning strategy, flow shop scheduling
PDF Full Text Request
Related items