Fast real-time SDRE controllers using neural networks

Rômulo Fernandes da Costa; Osamu Saotome; Elvira Rafikova; Renato Machado

doi:10.1016/j.isatra.2021.02.019

Fast real-time SDRE controllers using neural networks

ISA Trans. 2021 Dec:118:133-143. doi: 10.1016/j.isatra.2021.02.019. Epub 2021 Feb 16.

Authors

Rômulo Fernandes da Costa¹, Osamu Saotome², Elvira Rafikova³, Renato Machado⁴

Affiliations

¹ Graduate Program in Electronic and Computer Engineering - Electronic Devices and Systems, Electronic Engineering Division, Aeronautics Institute of Technology (ITA), 50 Praça Marechal Eduardo Gomes, São José dos Campos, SP, 12228-900, Brazil. Electronic address: Elromulo2006@yahoo.com.br.
² Department of Applied Electronics, Electronic Engineering Division, Aeronautics Institute of Technology (ITA), 50 Praça Marechal Eduardo Gomes, São José dos Campos, SP, 12228-900, Brazil. Electronic address: osaotome@ele.ita.br.
³ Engineering, Modeling and Applied Social Sciences Center, Federal University of ABC (UFABC), 5001 Av. dos Estados, Santo André - SP, 09210-580, Brazil. Electronic address: elvirarafikova@ufabc.edu.br.
⁴ Department of Telecommunications, Electronic Engineering Division, Aeronautics Institute of Technology (ITA), 50 Praça Marechal Eduardo Gomes, São José dos Campos, SP, 12228-900, Brazil. Electronic address: renatomachado@ieee.org.

PMID: 33612275
DOI: 10.1016/j.isatra.2021.02.019

Abstract

This paper describes the implementation of fast state-dependent Riccati equation (SDRE) control algorithms through the use of shallow and deep artificial neural networks (ANN). Several ANNs are trained to replicate an SDRE controller developed for a satellite attitude dynamics simulator (SADS) to display the technique's efficacy. The neural controllers have reduced computational complexity compared with the original SDRE controller, allowing its execution at a significantly higher rate. One of the neural controllers was validated using the SADS in a practical experiment. The experimental results indicate that the training error is sufficiently small for the neural controller to perform equivalently to the original SDRE controller.

Keywords: Deep learning; Neural control; SDRE control; Satellite attitude control; Stacked denoising autoencoders.