Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
Support Vector Regression based Nonlinear Model Predictive Control on FPGA
Submission Author:
Renato Sampaio , DF
Co-Authors:
Carlos Eduardo Silva Santos, CARLOS H Llanos, Ricardo Jacobi, Leandro dos Santos Coelho, Helon Vicente Hultmann Ayala, Renato Sampaio
Presenter: Renato Sampaio
doi://10.26678/ABCM.COBEM2017.COB17-1772
Abstract
Model based predictive control (MPC) is a control technique currently applied in industry which yields very effective control performance. Since it requires solving an optimization problem at each sampling interval, its computation cost is very high when compared to other control techniques. Recently, there is a great focus on applying MPC in real-time embedded systems. In the last decade various methods to achieve fast embedded solutions for Linear MPC were proposed. For more complex nonlinear systems though the problem remains open. This paper proposes a very fast hardware architecture for Nonlinear MPC by approximating an offline solution to a Support Vector Regressor (SVR) and implementing it on a FPGA. The SVR training process is done by the proposed Nature Inspired Optimization Tools for SVM (NIOTS) where various input configurations are explored in search of an optimal solution. The NIOTS tool is based on a Multi-Objective Particle Swarm Algorithm (MOPSO) and is able to generate a set of solutions that balance complexity and precision. The results show three SVR implementations in FPGA based on floating-point operations that are able to compute a control action in two about microseconds.
Keywords
nmpc, fpga, svr
