A Model of Neural Networks Fractals applied in RS codes (255, k) for R-IEDs on FPGAs

Abstract

In this research, a neural network was developed, designed for complex dynamic systems, in reconfigurable hardware. A Reed Solomon coding system (255, k) was selected as a case study. The design method consisted in the description of the equations in VHDL language and the identification of the correspondence between the stages of the neural circuit. Among the results, we have identified the fractal architecture, considered in the design of the encoders and their self-similar components, as well as the training of a decoding network, with parallel processing. An optimization of the circuit is obtained. A theoretical contribution was achieved, with the concept of FNN fractal neural networks, fractal design technique and concurrent neural models, which are not presented in previous studies. Likewise, practical advances are obtained with the optimization of hardware resources and energy efficiency. The correspondence of these circuits has been interpreted with advanced adaptive control and communication for R-IEDs, which constitutes a significant contribution for smart grid applications, under criteria of hardware re-usability.