A scalable deep reinforcement learning approach for minimizing the total tardiness of the parallel machine scheduling problem

Abstract

Various problems in the logistics field can be modeled as parallel machine scheduling problem (PMSP), which involves the optimized assignment of a set of jobs to a collection of parallel machines. Deep reinforcement learning (DRL) has demonstrated promising capability in solving similar problems. To this motivation, we propose a practical reinforcement learning-based framework to tackle a PMSP with new job arrivals and family setup constraints. We design a variable-length state matrix containing information of all jobs and employ a Recurrent Neural Network (RNN) model to represent the DRL agent. In the numerical experiment, we first train the agent on a small PMSP instance with 3 machines and 30 jobs. Then we implement this trained agent to solve a set of instances with significant larger instance. Its performance are also compared with two dispatching rules. The extensive experimental results demonstrate the scalability of our approach and its effectiveness across a variety of scheduling scenarios.