Application of Paraconsistent Neural Networks to investigate synergism in Combustion Systems Marcos Carneiro Rodrigues, João Inácio da Silva Filho

Abstract

Artificial Neural Networks are brain-inspired models that recognize complex patterns from data and can handle nonlinear problems. Paraconsistent logic manages contradictory information, being useful in situations of inherent uncertainty. Paraconsistent Neural Networks extend traditional models by incorporating paraconsistent logic principles. This study explores the capacity of these networks to analyze interactions between variables. A Paraconsistent Neural Network for 2 variables (with 5 inputs) is developed using MATLAB®, employing the LM method for training. The network achieves a Mean Squared Error (MSE) of 1.1x10-3 compared to a Sigmoid Neural Network (MSE of 4x10-3). The learned weights between x₁x₂ and H₁ indicates synergistic effect among input variables.