VOCAL 2024
Abstract Submission

76. Improvements of the Q-compression method for constrained stochastic graph traversal problems

Invited abstract in session TD-3: Stochastic optimization and applications I, stream Stochastic optimization and applications.

Thursday, 14:45 - 16:15
Room: C 104

Authors (first author is the speaker)

1. Tamás Kegyes
Department of Computer Science and Systems Technology, University of Pannonia
2. Alex Kummer
Department of Process Engineering, University of Pannonia
3. Zoltán Süle
Department of Computer Science and Systems Technology, University of Pannonia
4. János Abonyi
Department of Process Engineering, University of Pannonia

Abstract

We identified a diverse set of real-life problems belonging to the class of constrained stochastic graph traversal problems. Optimal routing problem of a truck, where the constraints describe the driver’s working hours limits and the parking availability belongs to the problem class as well as daily route planning of an electric delivery van, where the constraints are based on the limits of battery capacity and charging options, or disassembly line optimization problem, where a precedence graph describes component removal dependencies and the constraints limit the use of parallel workstations. The common root of these problems is that they lead to a sequential decision problem with a mixed-integer state space.
Hence, the classical solution methods are not directly applicable efficiently to the problem class, we turned to reinforcement learning methods. We developed a generally applicable Q-compression method for solving mixed-integer graph optimization problems, which is based on a dynamic discrete representation of the mixed-integer state space and provides a human-interpretable solution to the curse of dimensionality issue. The usability of our method is demonstrated in selected use cases of constrained stochastic graph traversal problems.
We are improving the Q-compression method by fine-tuning the framework’s discretization method, identifying further problem types for which our method is applicable, and parallelizing the training process for multiple agents.

Keywords

Status: accepted


Back to the list of papers