Collective Learning for Energy-centric Flexible Job Shop Scheduling

Abstract

Manufacturing industries can reduce their energy consumption by exploiting the flexibility in manufacturing processes, such as machine availability, flexible jobs, and resource usage. In this paper, we exploit the flexibility inherent in some jobs, their schedules, and their energy consumption to model an energy-centric flexible job shop scheduling problem. We assume that there is limited electrical energy to power machines, and we attempt to match the schedules of machine jobs to the available energy with the objective of minimizing the energy consumption. We propose that collective learning, i.e., a form of decentralized (and unsupervised) learning where autonomous agents coordinate their decision-making to collectively learn and manage tasks that can be efficiently performed by coordination, can be employed to achieve this. We present a methodology that combines a plangeneration algorithm with a collective-learning tool—Iterative Economic Planning and Optimized Selections (I-EPOS)—to solve this problem with near optimal-solutions. We apply the methodology to a practical dataset comprising 3 machines and 12 jobs and show that the energy consumption decreases by approximately 5% when they choose and schedule the jobs, instead of acting independently or without any coordination. We also show that it is possible to scale this method to a large number of jobs to obtain reasonable solutions quickly, and that coordination always outperforms uncoordinated and independent actions by increasing the energy savings.