Font Size: a A A

Study And Application Of Multi-objective Optimization Evolutionary Algorithm

Posted on:2020-05-11Degree:MasterType:Thesis
Country:ChinaCandidate:Y WangFull Text:PDF
GTID:2428330596975221Subject:Mechanical engineering
Abstract/Summary:PDF Full Text Request
Multi-objective optimization problems frequently occur in fields of scientific research and engineering applications,such as control system design,industrial scheduling,software engineering and resource allocation.These practical problems often involve multiple targets that are mutually crowded.How to find the optimal solution when multiple targets are related to each other is especially important,which makes multi-objective optimization became a popular research direction.The evolutionary algorithm is inspired by Darwin's theory of evolution.Because it does not need the prior information of the optimization problem,it can find multiple non-inferior solutions through spatial random search,thus become an important branch of multi-objective optimization design,which variously focus and application on many scientific research fields.The advantages and disadvantages of multi-objective evolutionary algorithms depend largely on the diversity and convergence of algorithms.At the same time,how to balance the exploration and development capabilities of algorithms has become a major difficulty in multi-objective evolutionary algorithms.Considering the above two aspects the main research contents include:(1)For improving the diversity of multi-objective optimization algorithms,a concentration-based algorithm is proposed based on the NSGA-III algorithm.In the non-dominated sorting process,the concentration information is obtained according to the similarity between the individual and the reference points.According to the individual concentration information,the probability of the individual entering the next generation is adjusted,which effectively avoid local optimum that caused by the assembling of a large number of similar non-inferior solutions.The algorithm has good diversity simultaneously ensuring the convergence of the algorithm.The convergence of the algorithm is verified by comparison with the real Pareto front end of the ZDT series and other test functions.At the same time,compared with other advanced algorithms,the performance index proves that the algorithm has improved in diversity.(2)Balanced the exploration and exploitation of multi-objective optimization algorithms,which means the local search and global search capabilities.The adaptive selection strategy is used to adjust the search method in the decision space,and the efficient and dynamic operation operator is used to update the next generation.The adaptive composition operator is proposed to balance the exploratory and development of the algorithm.The Simplex Crossover,the Simulated Binary Crossover and the Directional Crossover are combined as a crossover operator pool.Based on the performance of these operators,their recent fitness improvement rate is stored,and the operator is adaptively found to be more suitable for the next generation according to the performance of the previous generation.The operator then performs population updates and uses adaptive control strategies to adjust the parameters in each composite operator pool.The operator then performs a population update.The solution of UF series test function and comparison with other advanced algorithms show that the proposed algorithm has good robustness and convergence.(3)Multi-objective optimization design of permanent magnet drive.Based on the finite element three-dimensional model,the influence of structural variable parameters on output torque and eddy current loss is analyzed.The increase of output torque and the reduction of eddy current loss are taken as the optimization targets.The response surface method is used to establish the optimization model,and combined with the optimization algorithm proposed above.Solving,the calculation results show that the optimized driver performance is improved,which also confirms the superiority of the proposed algorithm.
Keywords/Search Tags:Multi-objective evolutionary algorithm, Non-dominated sorting, Diversity, Adaptive selection, Permanent magnet drive
PDF Full Text Request
Related items