Optimizing Resource Allocation Policies in Real-World Business Processes Using Hybrid Process Simulation and Deep Reinforcement Learning

Resource allocation refers to the assignment of resources to activities for their execution within a business process at runtime. While resource allocation approaches are common in industries such as manufacturing, directly applying them to business processes remains a challenge. Recently, techniques like Deep Reinforcement Learning (DRL) have been used to learn efficient resource allocation strategies to minimize the cycle time. While DRL has been proven to work well for simplified synthetic processes, its usefulness in real-world business processes remains untested, partly due to the challenging nature of realizing accurate simulation environments. To overcome this limitation, we propose DRL_HSM that combines DRL with Hybrid simulation models (HSM). The HSM can accurately replicate the business process behavior so that we can assess the effectiveness of DRL in optimizing real-world business processes. We evaluate our method on four real-world and two elaborate synthetic business processes, constrained by temporal resource availability and a restricted number of resources. An empirical evaluation shows that DRL_HSM outperforms the benchmarks by, on average, 45%, up to 307%, in 14 out of 24 considered evaluation scenarios and is competitive with the best-performing benchmark in 8 scenarios.