Michał Ostapowicz
supervisor: Piotr Gawrysiak
Assigning resources during the execution of business processes is a repetitive task that can be effectively automated. However, various automation methods may give suboptimal results. Proper resource allocation is vital as it leads to significant cost reductions and increased effectiveness, resulting in increased revenues. During the talk, an original representation that allows a multi-process environment modeling with different process-based rewards will be presented along with the simulation engine developed specifically for this purpose. Considered business processes usually share resources that differ in their eligibility.
This work will propose a method utilizing double deep reinforcement learning for an online resource allocation for multiple-process and multi-resource environments. Due to the usage of tabular algorithms, previous approaches using data in the form of event logs or applying online learning were limited by the exploding computational complexity when the number of possible states increased.
This presentation also shows the entire experimental setup. It compares results achieved using reinforcement learning and two popular strategies, namely Shortest Processing Time (SPT) and First-In-First-Out (FIFO), widely used in the industry. As a presentation summary, It is planned to show plans for future research, e.g., comparing reinforcement learning with more complex heuristics such as Monte Carlo tree search (MCST).