TB6 - Scheduling 3

Production scheduling in integrated process industries is a challenging optimization problem due to strong interactions among subsystems, multiple conflicting objectives, and complex operational constraints. Scheduling decisions in such processes must account for demand fulfillment, losses during product change transition periods, and energy consumption within highly nonlinear and dynamic environments.
This work proposes a simulation-based optimization framework leveraging Deep Reinforcement Learning (DRL) to address this problem without requiring explicit mathematical formulations. A Double Deep Q-Network (DDQN) agent is trained through repeated interactions with a process simulator, enabling the learning of adaptive scheduling policies directly from system responses. This approach alleviates the need for detailed explicit mathematical formulations of the optimization problem, which are often difficult to develop and maintain in such complex settings.
The proposed method is evaluated on an industrial-scale case study of a pulp and paper mill comprising several interconnected equipment, including pulp production equipment, a paper machine, evaporators, and boilers. Results demonstrate that the learned policy satisfies production demand while reducing product transition losses and improving energy efficiency by up to 8% compared to baseline operational rules, resulting in a measurable reduction in operating costs. These findings highlight the potential of DRL as a flexible and scalable optimization tool for large-scale scheduling problems, particularly in contexts where traditional model-based approaches are difficult to implement.

We propose a multi-objective framework for the optimal dispatch of aggregated distributed energy resources (DERs) to achieve distribution peak shaving while accounting for additional operational objectives. Over a long-term timescale, the framework determines the weights of these objectives through a multi-criteria decision-making procedure that captures the relative importance among objectives. For real-time implementation, DER dispatch setpoints are continuously computed using receding-horizon multi-objective robust optimization, in which an affine policy is employed to compensate and disaggregate uncertainties in load and renewable generation. The performance of the proposed framework is validated through both numerical case studies and real-time hardware-in-the-loop testing.

This paper studies a scheduling and queueing problem for electric multirotor drones in aerial thinning. Drones share limited operators and charging stations, creating queues that impact timing and battery usage. We propose a construction heuristic that coordinates task assignment and charging decisions to manage resource contention and reduce waiting times.
Keywords: drone scheduling; queueing; resource allocation; aerial thinning; construction heuristic.

We study a large-scale multi-project scheduling problem motivated by audit firms, where activities must be assigned to qualified resources over long planning horizons while accounting for workload overload and capacity constraints. The problem is formulated as a mixed-integer programming model with multiple overload-related objectives, including the minimization of maximum and total overload over different horizons.
To address the computational challenges posed by real-world instance sizes, we propose a fix-and-optimize matheuristic that decomposes the decision space and enables scalability. Computational experiments are conducted on real industrial data provided by a software company, involving annual planning instances with tens of thousands of activities.
We analyze the impact of alternative overload objectives under varying horizons and continuity requirements. Results show that min–max objectives effectively limit extreme overload peaks and promote workload fairness, whereas total-overload minimization concentrates workload on fewer employees and reduces resource usage. Furthermore, allowing limited flexibility in resource-activity assignments leads to substantial performance gains: reassigning only 5–25% of activities nearly eliminates overload.
These findings highlight fundamental trade-offs between fairness, flexibility, and efficiency, and demonstrate the value of the proposed framework as a decision-support tool for workload planning in professional service firms.

The Batch Pan Scheduling Problem represents a major operational challenge in white sugar refineries, where production planning must balance throughput objectives against high and fluctuating energy demands driven by crystallisation. This paper examines batch pan scheduling from an integrated production and energy-aware perspective. A discrete-time mixed-integer linear programming model and a mixed-integer multi-objective formulation are developed to represent the initiation of batch crystallisation cycles across sequential pan stages while accounting for total energy consumption. The models capture material flow dependencies, stage capacity constraints, and the temporal structure of industrial sugarhouse operations. The single-objective model maximises the number of completed batch cycles at the final pan stage within a given planning horizon, directly representing final sugar production. Building on this solution, the multi-objective model incorporates the achieved production level as a constraint with controlled relaxation to minimise total energy consumption. Numerical experiments investigate the influence of planning horizon length on scheduling flexibility and energy performance. Results indicate that extending the planning horizon enhances temporal coordination, enabling a more even distribution of operations and yielding substantial reductions in energy consumption without compromising production output. Furthermore, the multi-objective formulation achieves consistently stable energy performance, with total energy requirements comparable to or lower than those obtained from the single-objective model under the same horizon length. These findings demonstrate the critical role of planning horizon selection in energy-aware batch scheduling and show that energy savings and variability reduction can be realised through improved temporal coordination. The proposed framework provides a transparent and effective decision-support tool for energy-efficient production planning in batch process industries.