Publicação

Reducing the number of objectives for many-objectives optimization: empirical analysis of a machine learning approach

Ver documento

Detalhes bibliográficos
Resumo:The practical need of solving real-world optimization problems is faced very often of dealing with many objectives, but from the beginning, a question arises: Are all the objectives really necessary? The answer to this question lies in the complex relations existing between the parameters of the process, i.e., not only between the objectives and the decision variables (DVs), but also between the DVs and DVs and between the objectives and objectives. Simultaneously, intense research is made to improve the performance of multi-objective population-based algorithms to deal with many objectives that, often, imply complex algorithms and time consuming computations with complex results that experts on the field of the problem might not understand and, as a consequence, did not accept and apply in practice. A straightforward alternative is to infer the complex relations between the process parameters with the aim of reducing the number of objectives. The use of Machine Learning (ML) methodologies for that can be very useful since it is someway demonstrated in the literature on the subject of reducing the number of objectives. In this work, ML is used to reduce the number of objectives and the results are assessed empirically using a real-world application.
Autores principais:Gaspar-Cunha, A.
Outros Autores:Costa, Paulo; Monaco, Francisco; Delbem, Alexandre
Assunto:Multi-objective optimization Machine Learning Evolutionary Algorithms Reduction of the number of objectives
Ano:2023
País:Portugal
Tipo de documento:capítulo de livro
Tipo de acesso:acesso aberto
Instituição associada:Universidade do Minho
Idioma:inglês
Origem:RepositóriUM - Universidade do Minho
Descrição
Resumo:The practical need of solving real-world optimization problems is faced very often of dealing with many objectives, but from the beginning, a question arises: Are all the objectives really necessary? The answer to this question lies in the complex relations existing between the parameters of the process, i.e., not only between the objectives and the decision variables (DVs), but also between the DVs and DVs and between the objectives and objectives. Simultaneously, intense research is made to improve the performance of multi-objective population-based algorithms to deal with many objectives that, often, imply complex algorithms and time consuming computations with complex results that experts on the field of the problem might not understand and, as a consequence, did not accept and apply in practice. A straightforward alternative is to infer the complex relations between the process parameters with the aim of reducing the number of objectives. The use of Machine Learning (ML) methodologies for that can be very useful since it is someway demonstrated in the literature on the subject of reducing the number of objectives. In this work, ML is used to reduce the number of objectives and the results are assessed empirically using a real-world application.