PSE-2021

A RECEDING HORIZON-ORIENTED AUTO-TUNING FRAMEWORK FOR MPC STRATEGIES

Seminários Process System Engineering - PSE UFBA

Prof. Raony M. FontesDEQ - Escola Politénica Universidade Federal da Bahia

28/04/2021

Introduction

Establishing the context, background, and importance of the topic

https://pubs.acs.org/doi/10.1021/acs.iecr.9b03502

Introduction

Establishing the context, background, and importance of the topic

Model Predictive Control

MPC

Different prediction models, objective functions, process constraints.

Flexible control strategy

Optimizer

Process

Non-intuitive interaction between the parameters

Tuning

Model

Introduction

Establishing the context, background, and importance of the topic

Objectives

General

To present a novel MPC auto-tuning framework based on a receding optimization problem, which is flexible in applying to different MPC formulations, different tuning criteria, and designed for online implementation.

Objectives

Contributions

General

To present a novel MPC auto-tuning framework based on a receding optimization problem, which is flexible in applying to different MPC formulations, different tuning criteria, and designed for online implementation.

To address the proposed framework flexibility in the face of different MPC strategies

To design a flexible receding horizon optimization framework

To design a monitoring and tracking layer suitable for different tuning requirements

Proposed framework

A method based on optimization with the capacity for online applications which is flexible for different MPC

Proposed framework

A method based on optimization with the capacity for online applications which is flexible for different MPC

Automatic Layer

Responsible for providing the simulated values of the controlled and manipulated variables

Monitoring

Evaluates the new parameters in order to meet the desired performance criteria

Tuning

Proposed framework

A method based on optimization with the capacity for online applications which is flexible for different MPC

MPC

Auto-tuning

Receding horizon optimization

Receding horizon optimization

Proposed framework

A method based on optimization with the capacity for online applications which is flexible for different MPC

MPC

Auto-tuning

Receding horizon optimization

Receding horizon optimization

Open-loop modelControl actions Prediction horizonControl horizon

Closed-loop simulationTuning actionsSimulation horizonTuning horizon

Info

Info

Proposed framework

A method based on optimization with the capacity for online applications which is flexible for different MPC

Results

Testing the framework for different formulations of model predictive control and its different scenarios of fit criteria

Results

Testing the framework for different formulations of model predictive control and its different scenarios of fit criteria

Tuning DMC Binary column

DMC

To represent the model uncertainties, the gains are assumed at ±20%, time constants ±10% and time delay ±10

Linear mismatch

Optimizer

Process

Model

Tuning

Tuning actions consist of relative increments on the matrix elements of outputs weights (Q) and move suppression (R)

Results

Testing the framework for different formulations of model predictive control and its different scenarios of fit criteria

Results

Testing the framework for different formulations of model predictive control and its different scenarios of fit criteria

Tuning IHMPC CSTR

IHMPC

Optimizer

The process is represented by a set of nonlinear ordinary differential equations

Nonlinear mismatch

Process

Tuning

Tuning actions consist of relative increments on the matrix elements of outputs weights (Q) and move suppression (R)

Model

Results

Testing the framework for different formulations of model predictive control and its different scenarios of fit criteria

Final considerations

Summarizing the preliminary contributions of this work and suggests the future stages.

Final consideration

Summarizing the preliminary contributions of this work and suggests the future stages.

Conclusions

The proposed auto-tuning framework features set out to supply the lack of flexible tuning methods for different MPC formulations. So, the findings from this study make meaningful contributions to the current tuning MPC literature.

Final considerations

Summarizing the preliminary contributions of this work and suggests the future stages.

Remarks and suggestions

Conclusions

The proposed auto-tuning framework features set out to supply the lack of flexible tuning methods for different MPC formulations. So, the findings from this study make meaningful contributions to the current tuning MPC literature.

The framework provides greater flexibility for applications regarding MPC type, process type, performance requirements.

The framework identifies nonconformities by a mismatched closed-loop simulation some steps time ahead. From that point, it triggers the computing of new optimal parameters.

Thanks!

Raony Maia Fontes

Professor Departamento de Eng. QuímicaEscola Politécnica - UFBA

raony@ufba.br

http://lattes.cnpq.br/9115757149952281