Proposing a Reinforcement Learning Algorithm for Optimizing Human Resource Allocation in Municipal Service Units (Case Study: Ramhormoz Municipality)
Proposing a Reinforcement Learning Algorithm for Optimizing Human Resource Allocation in Municipal Service Units (Case Study: Ramhormoz Municipality)
Hossein Maleki1
1) B.Sc. in Computer Software Engineering, Islamic Azad University, Mahshahr Branch; M.Sc. in Geography and Urban Planning, Urban Environmental Specialization, Payame Noor University, Ahvaz Branch, Iran
IOI Code
:
ABEB-ACACDC
Publication :
4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
The present study was conducted with the aim of proposing a reinforcement learning algorithm for the dynamic optimization of human resource allocation in the service units of Ramhormoz Municipality. The research method was applied, using a mixed qualitative and quantitative approach. In the qualitative section, fifteen semi-structured interviews were conducted with experts and managers of the service units, and six main themes were extracted from 127 initial codes. In the quantitative section, the problem was modeled as a Markov Decision Process, and the proximal policy optimization algorithm with an actor-critic architecture was implemented in a simulated environment consisting of six service units and 100 staff members. An action-masking mechanism was used to ensure compliance with operational constraints. The findings showed that the reinforcement learning algorithm, compared with the rule-based method, achieved a 50 percent reduction in average service waiting time (from 22.7 to 11.4 minutes), a 12.6-percentage-point increase in the demand fulfillment rate (from 81.6 to 94.2 percent), and a reduction in the coefficient of variation of workload from 0.34 to 0.17. This finding is consistent with the study by Li (2024), which raised the task completion rate to 93.6 percent; however, unlike the intersection traffic control study, the superiority of reinforcement learning over the static method was not evident under stable conditions and was observed only under dynamic conditions involving sudden changes in demand. The main limitations of the study are that the results are based on simulation and that the model is single-agent. For future research, an operational pilot implementation, extension of the model to a multi-agent approach, and the addition of citizen and staff satisfaction indicators are recommended. In conclusion, reinforcement learning, as an intelligent decision-support tool within a human decision-making framework, can significantly enhance the efficiency of limited resources in small Iranian municipalities and improve service delivery.
Keywords : human resources reinforcement learning ramhormoz municipality dynamic optimization urban resource management