본문 시작

본문

Online Workshop Website [link]

** You can log-in with your ID and PWD used for conference registration site ([link]).

** Additional Zoom PWD will be sent to registered email addresses at Noon on 6/13 (Mon, KST).

Registration for Workshop 
** The workshop registration site opened, https://dycops2022workshop.sigongji.com/
** Workshop 3 is cancelled.
**  No refunds will be made, and are given for No-shows.

 Half Day

Regular 

100 EUR

Student 

  70 EUR 

 Full Day 

Regular 

 150 EUR 

Student 

 100 EUR 


All Workshops on June 14th, 2022 (Tuesday) (Workshop 3 is cancelled)

 Workshop 1: Machine learning for predictive control and real-time optimization 

 Antonio del Rio-Chanona, Panagiotis Petsagkourakis, Benoit Chachuat (UK) 

(Switched to Full-Virtual (Live zoom in Room Rose & zoom), FULL-DAY, 10:00AM-6:30PM)

 Brief Introduction

 The first part of the workshop will be dedicated to reinforcement learning (RL) for optimization and control. The emerging field of RL has led to remarkable empirical results in rich data domains like robotics and strategy games. However, so far, no adoption has been made into process engineering. This workshop aims to introduce and showcase the use of RL in process control and operations. An introduction to RL will initially be given in the first talk, where different methods will be conceptually explained and preliminary mathematical formulations will be explored [4]. The second talk will highlight the recent developments on how models can efficiently be used for safe and fast learning of RL agents in practical implementations [5,6,7].


The second part of this workshop will be concerned with real-time optimization (RTO), with a view to enhancing RTO using supervised learning techniques. The supervised learning method of choice, Gaussian process (GP) regression [1], will be reviewed in the first talk alongside Bayesian optimization. The second talk will present a brief summary of different RTO paradigms, with particular emphasis on modifier adaptation [2]. A recent methodology whereby GP regression is integrated within modifier adaptation in combination with trust-region and Bayesian optimization approaches [3] will be presented in the third talk. The last talk in this part of the workshop will highlight a promising development that exploits multi-fidelity GP regression to further improve the convergence rate and practical implementation of GP-assisted RTO.


[1] Rasmussen, C. E., & Williams, C. K. I. , Gaussian Processes for Machine Learning.

[2] A. Marchetti, B. Chachuat, and D. Bonvin, Modifier-Adaptation Methodology for Real-Time Optimization, Industrial & Engineering Chemistry Research, 2009, 48 (13), 6022-6033

[3] E. A. del Rio Chanona, P. Petsagkourakis, E. Bradford, J. E. Alves Graciano, B. Chachuat, Real-time optimization meets Bayesian optimization and derivative-free optimization: A tale of modifier adaptation, Computers & Chemical Engineering, 2021, 107249 (147)

[4] Andrew Barto and Richard S. Sutton, Reinforcement Learning: An Introduction

[5] P. Petsagkourakis, I.O. Sandoval, E. Bradford, D. Zhang, E.A. del Rio-Chanona, Reinforcement learning for batch bioprocess optimization, Computers & Chemical Engineering, 2020, 106649 (133)

[6] Max Mowbray, Panagiotis Petsagkourakis, Ehecatl Antonio del R´io Chanona, Robin Smith, Dongda Zhang, Safe Chance Constrained Reinforcement Learning for Batch Process Control, 2021, https://arxiv.org/abs/2104.11706

[7] Distributional Reinforcement Learning for Scheduling of Chemical Production Processes, Max Mowbray, Dongda Zhang, Ehecatl Antonio Del Rio Chanona, https://arxiv.org/pdf/2203.00636.pdf 


 10:00 - 11:15

 Reinforcement Learning introduction (virtual) - Antonio

 11:15 - 12:30

 Reinforcement learning for control and operations research (virtual) - Antonio

 

 Lunch Break

 2:00 - 3:30

 Gaussian processes and Bayesian Optimization (virtual) - Panos

 3:30 - 4:30

 Real-time optimization (virtual) - Benoit

 4:30 - 5:30

 Real-time optimization with Machine Learning (virtual) - Benoit

 5:30 - 6:30

 Real-time optimization with Mutifidelity Gaussian processes (virtual) - Panos



 Workshop 2: InfiniteOpt.jl: A Julia Package for Infinite-Dimensional Optimization 

Joshua Pulsipher (USA) 

(HYBRID (Room Lilac & zoom), FULL-DAY, 9:00AM-5:00PM, Laptop is needed for practice)

 Brief Introduction

In this workshop, we will address how to use InfiniteOpt.jl to model/solve nonlinear optimal control problems and how it enables novel formulations (e.g., using stochastic risk measures to shape dynamic cost trajectories). InfiniteOpt.jl is a Julia-based open-source software package (an extension of the popular mathematical programming modeling package JuMP.jl) that provides an intuitive symbolic interface to compactly model a wide breadth of problem classes which include dynamic, PDE-constrained, and stochastic optimization problems (and combinations thereof). Moreover, it is built modularly such that advanced users can quickly extend it to implement their cutting-edge modeling/solution techniques to make them accessible to a wide audience of individuals with a limited technical background. All these aspects make InfiniteOpt.jl a powerful tool for both practitioners and advanced researchers alike in tackling advanced optimal control problems.


Details can be found [here] 


 Workshop 3: Constrained-based Modelling Applications in Bioprocesses (Cancelled)

Rudiyanto Gunawan, Dong-Yup Lee, Steffen Waldherr, Hyun-Seob Song (USA)

  (HYBRID (Room Cosmos & zoom), HALF-DAY, 1:00PM-5:00PM)

 Brief Introduction

 The goal of the workshop is to give an overview of constrained-based modeling (CBM) and highlight various applications of CBM in metabolic engineering and biomanufacturing.


 We start with a brief overview of the history and recent progress of CBM that center around flux balance analysis, the curation of genome-scale metabolic models, and their various applications in biomanufacturing settings. Next, we highlight computational tools for CBM, where FBA will be a major focus. We continue with methods for integrating omics data with CBM and how these methods enable omics-guided metabolic engineering. We further provide a review of state-of-the-art modelling and in silico tools for microbiome design. Finally, control-oriented applications of CBM in biomanufacturing are presented, including observer and controller designs and model predictive control of bioprocesses. 


Details can be found [here].



 Workshop 4: Litium-Ion Batteries: Challenges and Opportunities

 Ulrike Krewer, Richard D. Braatz, Jeesoon Choi, Rolf Findeisen, Jay H. Lee, Davide Raimondo

  (HYBRID (Room Tulip & zoom), 8:30AM-5:30PM, FULL-DAY)

 Brief Introduction

In this workshop, we will introduce the attendees to battery design, modelling and estimation, and optimization of battery operation. The emphasis will be on using modern modeling/parameter estimation methods in conjunction with machine learning and optimization tools.


Details can be found [here] (Davide Raimondo's presentation is cancelled)



 Workshop 5: The Next Epoch of Model Predictive Control: Exploiting Machine Learning Methods for Approximation and Design

 Joel Paulson, Ali Mesbah (USA)

  (Fully Virtual (zoom), 8:30AM-1:30PM, HALF-DAY) 

 Brief Introduction

This half-day workshop will consist of 5 separate presentations. The first part will provide a detailed introduction to the fields of machine learning and MPC, and clearly motivate the challenges with the current state-of-the-art in MPC (in the form of many different illustrative examples). This introduction will set the stage for the next 3 modules, each of which discusses very recent advances on combining specific machine learning methods with MPC to address one of the key challenges

defined in the introduction. Module 1 will focus on so-called learning-based MPC methods that can learn the plant-model mismatch from data, while accounting for uncertainty to ensure some degree of robustness. Module 2 will focus on the use of deep learning methods to approximate any arbitrarily complex MPC law. Module 3 will demonstrate how Bayesian optimization can be used as a flexible framework for learning an optimal set of tuning parameters for MPC laws in a data-efficient manner. An approximately twenty-minute code demonstration will be performed at the end of each module. The workshop will conclude with a final presentation that discusses some remaining open research problems in this area as well as some more detailed applications of the methods to

next-generation systems. All workshop notes and slides will be provided, and open-source codes for selected examples and applications will be made available on the workshop’s website.


Introduction: Why Go Beyond Nominal MPC? (45 minutes) 

Module 1: Learning Plant-model Mismatch (1 hour)

Module 2: Learning Efficient MPC Representations (1 hour)

Module 3: Learning Optimal MPC Tuning under Uncertainty (1 hour)

Applications to Next-Generation Energy and Manufacturing Systems (30 minutes)

Closing Remarks and Q&A (15 minutes)

* Note that each module will have a “code review” wherein we illustrate how the concepts can be applied in practice using example scripts in Matlab/Python.