Towards Tailoring Reinforcement Learning to Solve the Online Surgery-Planning and Scheduling Problem

Abstract

This paper addresses the online surgery planning and scheduling problem for operating rooms and recovery beds.We aim to minimize the makespan by dynamically assigning surgery dates, operating rooms, and recovery beds.Our integrated framework uses a Mixed-Integer Linear Program (MILP) solved with Python's PuLP package for initial scheduling and Reinforcement Learning (RL) for real-time adjustments.The MILP provides a static schedule, while RL handles disruptions and updates schedules using experience replay and target networks for stable training.Preliminary results demonstrate that this approach effectively manages scheduling complexities, improving operational efficiency and minimizing idle times.This highlights the potential of combining MILP and RL for adaptive surgical scheduling