An Integrated Stochastic Deep Learning–Short-Term Production Scheduling–Optimal Control Framework for General Batch Processes

Abstract

Integrated operational decision-making in chemical plants is important for improving profitability. Integrated scheduling and control frameworks have been developed to enhance coordination between tactical and operational decisions. As such frameworks start to incorporate more features (e.g., uncertainty, detailed nonlinear process models) the computation time and resources that they require may increase significantly and lead to solutions that cannot be implemented online. Leveraging deep learning models that can capture process nonlinearities without a significant computational burden is a potential solution to this problem. Motivated by the above, we propose a two-stage stochastic integrated deep learning−scheduling−optimal control framework that can be used to model any batch process and can be solved efficiently. The benefits are demonstrated using a prototype case study.

A machine learning framework for improving refinery production planning

“…The solution of problem (7) that stabilizes (reduces the variability) of the estimated parameters of all outputs is selected by using trace plots. 31 For standard linear regression in which a closed form equation to estimate model parameter exists, expressions to account for uncertainty in the model (variability of estimated parameters) and predictions are readily available.…”

“…31 For standard linear regression in which a closed form equation to estimate model parameter exists, expressions to account for uncertainty in the model (variability of estimated parameters) and predictions are readily available. 31 However, for formulation (7) in which a robust approach is considered and mass balance is enforced, there are no closed-form expressions that account for model uncertainty.…”

“…Over the last years, extensive research and several frameworks/algorithms have been proposed to integrate adjacent decision-making layers such as production planning and production scheduling, [1][2][3][4][5] production scheduling and process control, [6][7][8][9][10] and real-time optimization and (advanced) process control. [11][12][13][14][15] While these studies have demonstrated the value of integrated operation, only limited research has been carried out to devise computational tractable frameworks that integrate non-adjacent decision-making layers (e.g., production planning and process control) and that can be implemented in practice.…”

Editorial for the Festschrift Honoring the Legacy of Babatunde A. Ogunnaike (1956–2022)

“…5 Stengel et al discuss an interactive tool for designing solvent recovery systems. 6 Correlations between interfacial area and volumetric mass transfer in bubble columns are presented by Hazare et al 7 Special Issue: In Honor of Babatunde A. Ogunnaike Process control and operations are represented by the works of Santander et al, who discuss an integrated stochastic framework for simultaneous short-term scheduling and control for batch processes, 8 Lovelett et al, who review the control of processes with input multiplicity, 9 and Pinnamaraju et al, who employ sparse optimization to construct soft sensors from irregularly and infrequently sampled data. 10 Finally, randomness and the use of feedback control to address it are covered in the works of Pearson 11 and McAllister et al, 12 respectively.…”

“…Process control and operations are represented by the works of Santander et al, who discuss an integrated stochastic framework for simultaneous short-term scheduling and control for batch processes, Lovelett et al, who review the control of processes with input multiplicity, and Pinnamaraju et al, who employ sparse optimization to construct soft sensors from irregularly and infrequently sampled data . Finally, randomness and the use of feedback control to address it are covered in the works of Pearson and McAllister et al, respectively.…”