(61n) Integration of Production Planning and Scheduling Problems with Uncertainty and Feasibility Analysis

Production planning and scheduling are two important problems in chemical production operations. Production planning aims at satisfying the production targets based on demand profile over a mid-term horizon, while production scheduling focuses on outlining the optimal task sequences following the production planning over a short-term horizon¹. The integrated optimization of the two levels avoids infeasible solutions caused by overestimating production capacity of scheduling level, and therefore promotes production efficiency. Additionally, as the demand volatilities of chemicals and energy supply fluctuations, the decision making process has to improve by considering uncertainties. However the computational cost often becomes a significant burden for decision making, and machine learning integration has become a popular tendency to improve computational efficiency in this competitive market².

This presentation addresses simultaneous modeling and optimizing of production planning and scheduling problems with regard to demand uncertainty in planning level and utility uncertainty in scheduling level. In order to characterize the utility uncertainty more realistically, two kinds of utilities are classified: the utilities that are operated continuously (e.g., steam) and the utilities that have on/off characteristics (e.g., electricity). The volatility of supply and time-of-use electricity price are considered separately for the two utilities. Rolling horizon method is adopted to solve the integration problem and feasibility analysis will be executed to direct the planning level to generate more appropriate and computational tractable production targets for scheduling level.

We propose a MILP framework for batch production systems with a discrete-time formulation. Rolling horizon method solves the planning and scheduling problem with a sequence of iterations; the planning model are solved within the time-window and only the production targets of the first planning period are transferred to the scheduling level. The time-window will roll forward and update its initial state according to the information transferred by the scheduling level until it reaches the last period of planning horizon. For the scheduling model, two kinds of micro-periods which depicts production and utility supply separately are considered. This avoids the problem that the duration of utility disturbance is not synchronized with the production batch duration. Additionally, we will use machine learning approaches to identify the interactions between planning and scheduling feasible regions.

We will show that this strategy is advantageous with respect to the following interrelated aspects: (1) it maximizes the profit of the enterprise in an uncertain environment and provides the optimal production sequence and material flow; (2) it allows infeasible solutions for scheduling level when the production targets overestimate the capacity of scheduling level, while it also allows early production for future periods; (3) it demonstrates that we can maximize the profits to meet the demand while significantly reducing the energy cost; (4) it performs better when employing feasibility analysis. To the best of our knowledge, no research in chemical production scheduling takes the uncertainty of different utilities types into account, so this work will depict the uncertainty according to its characteristics, and benchmarks will be made to reveal the advantages of the uncertainty model.

References：

1. Beykal B, Avraamidou S, Pistikopoulos EN. Data-driven optimization of mixed-integer bi-level multi-follower integrated planning and scheduling problems under demand uncertainty. Comput Chem Eng. 2022;156:107551. doi:10.1016/j.compchemeng.2021.107551
2. Badejo O, Ierapetritou M. Integrating tactical planning, operational planning and scheduling using data-driven feasibility analysis. Comput Chem Eng. 2022;161:107759. doi:10.1016/j.compchemeng.2022.107759