Estudio computacional de algoritmos de optimización de primer orden para la aceleración del control predictivo basado en modelo (MPC) en FPGAs
Abstract: Many of the most recent publications in the embedded MPC field choose to use first-order algorithms as a solution to the complexity of accelerating second-order algorithms in systems with restricted computing resources. Thus, the purpose of this project consist on evaluating the response o...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2020 |
| País: | España |
| Institución: | Universidad del País Vasco |
| Repositorio: | Addi. Archivo Digital para la Docencia y la Investigación |
| OAI Identifier: | oai:addi.ehu.eus:10810/46930 |
| Acceso en línea: | http://hdl.handle.net/10810/46930 |
| Access Level: | acceso abierto |
| Palabra clave: | control predictivo algoritmos de optimización AMA ADMM método de Hessiana-reducida MPC embebido model predictive control optimization algorithms reduced-Hessian method embedded MPC kontrol prediktiboa optimizazio-algoritmoak Hesiar-murriztuaren metodoa MPC txertatua |
| Sumario: | Abstract: Many of the most recent publications in the embedded MPC field choose to use first-order algorithms as a solution to the complexity of accelerating second-order algorithms in systems with restricted computing resources. Thus, the purpose of this project consist on evaluating the response of these first order algorithms and compare their performance with second-order algorithms to have consistent criteria to design specific accelerators optimized in the development of SoCs with application to the high performance embedded MPC. To aim this objective, the first part of the text focuses on explaining a background needed to understand the approach of the optimization algorithms in the solution of a reformulated MPC problem arising from a state-space model. In this chapter, definitions such as Shift and Delta discretisation, System Variable and Dynamics Stacking, and Sparse and Dense formulation are explained. Once the related background is presented, it is analysed in detail the application to the MPC of the most current and efficient first order algorithms, making a study of the proposals published in recent years. In this section, definitions such as optimality, stopping conditions and convergence will be taken into account. After this analysis, the application of these algorithms will be adapted to different MPC formulation options where it will be evaluated their computational performance in different scenarios. Lastly, the results obtained of the previous evaluation will be compared with the performance of the second order algorithms, particularly using a simplified Interior Point which it is optimised for an embedded MPC implementation. This comparison will allow to take appropriate design decisions in the implementation of MPC in embedded systems considering aspects such as hardware simplicity, computational performance and numerical characteristics. |
|---|