Identifica\c{c}\~ao de Sistemas N\~ao Lineares Utilizando o Algoritmo H\'ibrido e Bin\'ario de Otimiza\c{c}\~ao por Enxame de Part\'iculas e Busca Gravitacional

TL;DR

This paper introduces a novel hybrid and binary particle swarm optimization algorithm combined with gravitational search to improve the selection of polynomial NARX model structures, balancing accuracy and interpretability.

Contribution

The work proposes a new meta-heuristic algorithm for polynomial NARX model structure selection, emphasizing penalization based on regressor contribution and demonstrating advantages over existing methods.

Findings

Outperforms Error Reduction Ratio in model selection.

Achieves better trade-off between accuracy and interpretability.

Validated on electromechanical and electric heater systems.

Abstract

This work presents a new meta-heuristic approach to model structure selection of polynomial NARX models. In this respect, the technique penalizes the models based on the individual contribution of each regressor in representing the system. The new algorithm is tested on two experimental case studies: the identification of an electromechanical system and a eletric heater. The results are compared with Error Reduction Ratio and another meta-heuristic approach. The proposed method shows its advantages over compared methods in terms of the trade-off between prediction accuracy and model interpretability. The results are quantified and compared using the Mean Squared Error (MSE) indices.