Support vector machines and particle swarm optimization applied to reliability prediction

AUTOR(ES)

Isis Didier Lins

DATA DE PUBLICAÇÃO

2009

RESUMO

Reliability is a critical metric for organizations since it directly influences their performance in face of the market competition, as well as is essential in maintaining their production systems available. The prediction of such quantitaive metric is then of great interest, as it may anticipate the knowledge about system failures and let organizations avoid and/or overcome such undesirable situations. Systems reliability depends on the inherent aging factors as well as on the operational conditions the system is subjected to. This may render the reliability modelling very complex and then traditional stochastic processes fail to accurately predict its behavior in time. In this context, learning methods such as Support Vector Machines (SVMs) emerge as alternative to tackle these shortcomings. One of the main advantages of using SVMs is the fact that they do not require previous knowledge about the function or process that maps input variables into output. However, their performances are affected by a set of parameters that appear in the related learning problems. This gives rise to the SVM model selection problem, which consists in choosing the most suitable values for these parameters. In this work, this problem is solved by means of Particle Swarm Optimization, a probabilistic approach based on the behavior of biological organisms that move in groups. Moreover, a PSO+SVM methodology is proposed to handle reliability prediction problems, which is validated by the resolution of examples from literature based on time series data. The obtained results, compared to the ones provided by other prediction tools such as Neural Networks (NNs), indicate that the proposed methodology is able to provide competitive or even more accurate reliability predictions. Also, the proposed PSO+SVM is applied to an example application involving data collected from oil production wells

ASSUNTO(S)

support vector machines previsão de confiabilidade support vector machines otimização engenharia de produção engenharia de producao confiabilidade otimização via nuvens de partículas

Documentos Relacionados

• The Prediction of the Man-Hour in Aircraft Assembly Based on Support Vector Machine Particle Swarm Optimization
• Profiled support vector machines for antisense oligonucleotide efficacy prediction
• Applying Support Vector Machines for Gene ontology based gene function prediction
• Fish swarm optimization algorithm applied to engineering system design
• Support vector machines na classificação de imagens hiperespectrais