Título: Modelos Locais Crescentes para Identificação de Sistemas com Outliers: Um Estudo Comparativo
Autores: Jéssyca A Bessa, Guilherme A Barreto, Darlan A Barroso
Resumo: In this paper, we compare evolving variants of the growing local models (RAN and INC-LLM) for recursive dynamical system identification. The proposed model has the following features: growing online structure, fast recursive updating rules, better memory use (no storage of covariance matrices is required) and outlier-robustness. In this regard, efficiency in performance and simplicity of implementation are the essential qualities of the proposed approach. The proposed outlier-robust versions of RAN and INC-LLM results from a synergistic amalgamation of two simple but powerful ideas. For this purpose, we combine the neuron insertion strategy and outlierrobust LMS-like parameter estimation rules. A comprehensive evaluation involving two benchmarking data sets corroborates the proposed approachs superior predictive performance in outlier-contaminated scenarios compared to original models.
Palavras-chave: Growing Local models, System identification, Least Mean Estimate
Páginas: 8
Código DOI: 10.21528/CBIC2023-044
Artigo em pdf: CBIC_2023_paper044.pdf
Arquivo BibTeX: CBIC_2023_044.bib