(59am) Self-Optimizing Control Methodology Using Surrogate Models for Complex Systems: A Jupyter-Based Application for Flexible Exploration and Adjustment

Designing control structures for chemical processes is a highly challenging task due to the complex and nonlinear nature of these systems. The optimization of chemical processes involves taking into account various operational, environmental, and economic constraints. To ensure optimal operation, control structures must be carefully designed to maintain process specifications within strict constraint limits, while also accounting for unmeasured or random disturbances. Despite the critical importance of such control structures, their design remains highly intricate and involves navigating through a maze of interdependent variables and nonlinearities. In the late 2000’s a significant number of investigations were performed to improve and investigate how to find the best set of controlled variables which, when kept at constant setpoints, indirectly lead to near-optimal operation with acceptable loss [1-6]. This systematic procedure was denoted as “Self-optimizing control” technology [7]. These previous works remarks that many complex chemical systems pose a significant challenge for self-optimizing control methodologies, as the evaluation of Hessians and gradients can become inaccurate due to non-linearities and complex dependencies.

To address this challenge, recent studies have proposed the use of surrogate models to simplify the mathematical procedures and reduce the evaluation time. Surrogate models have shown promising results in accurately estimating the Hessians and gradients required for self-optimizing control, thus allowing for the effective implementation of this technology in complex chemical systems. Then, the authors proposed the use of surrogate models to simplifying the evaluation of such mathematical procedures at a reduced time [8]. Moreover, the switching among the set of control variable can be avoided since we can obtain the Hessians and gradients estimation at any point, from the initial conditions to close to optimal operation point using the surrogate models. The results show a robustness and stability of systematic, once it always finishes with the Hessians and gradients calculation close to the optimal operation point, required for self-optimizing control, and independent from the initial conditions, disturbance distance and process nonlinearities. The methodology was implemented in a Python based application called “Metacontrol” [9].

This work uses the systematic procedure based on surrogate models for a self-optimizing control methodology [8] with all the codes implemented in a Jupyter Notebook to allow the user to explore the tool from different perspectives. Initially, a latin hypercube sampling is used to generate inputs to the chemical system, and all relevant outputs of this system are used to train a representative surrogate model of the process using Kriging interpolators. The automation of the Aspen Hysys Process Simulator is available to run the experimental design. Then the optimization of an economic objective function is performed to obtain the optimal point and the possible active constraints. Finally, the optimal linear combination of measurements is determined to generate a control structure that minimizes the loss resulting from disturbances and implementation errors. The methodology is evaluated with a reactor with an exothermic generic reaction and the results presented in this work reproduces the results found in the literature [5; 10]. To sum up, the structure proposed in the work allows the use and application of the self-optimizing control methodology in a way the user have total flexibility to explore and adjust its studies and analysis.

Keywords: self-optimizing control, surrogate model, kriging, Jupyter notebook

References

[1] Halvorsen, Ivar J., et al. "Optimal selection of controlled variables." Industrial & Engineering Chemistry Research 42.14 (2003): 3273-3284.

[2] Alstad, Vidar, and Sigurd Skogestad. "Null space method for selecting optimal measurement combinations as controlled variables." Industrial & engineering chemistry research 46.3 (2007): 846-853.

[3] Alstad, Vidar, Sigurd Skogestad, and Eduardo S. Hori. "Optimal measurement combinations as controlled variables." Journal of Process Control 19.1 (2009): 138-148.

[4] Kariwala, Vinay. "Optimal measurement combination for local self-optimizing control." Industrial & Engineering Chemistry Research 46.11 (2007): 3629-3634.

[5] Kariwala, Vinay, Yi Cao, and S. Janardhanan. "Local self-optimizing control with average loss minimization." Industrial & Engineering Chemistry Research 47.4 (2008): 1150-1158.

[6] Kariwala, Vinay, and Yi Cao. "Bidirectional branch and bound for controlled variable selection. Part II: Exact local method for self-optimizing control." Computers & chemical engineering 33.8 (2009): 1402-1412.

[7] Skogestad, Sigurd. "Plantwide control: The search for the self-optimizing control structure." Journal of process control 10.5 (2000): 487-507.

[8] Alves, Victor MC, et al. "Metamodel-based numerical techniques for self-optimizing control." Industrial & Engineering Chemistry Research 57.49 (2018): 16817-16840.

[9] Lima, Felipe Souza, Victor Manuel Cunha Alves, and Antonio Carlos Brandao Araujo. "Metacontrol: A Python based application for self-optimizing control using metamodels." Computers & Chemical Engineering 140 (2020): 106979.

[10] Alstad, V. (2005). Studies on selection of controlled variables. Norwegian University of Science and Technology.