The overhead crane scheduling problem has been of interest to many researchers. While most approaches are optimization-based or use a combination of simulation and optimization, this research suggests a combination of dynamic simulation and reinforcement learning-based AI as a solution.
The goal of this steel plant simulation project was to minimize the crane waiting time at the LD converters by creating a better crane schedule.
Firstly, an agent-based model of a plant was created in AnyLogic simulation software. Then, the modelers had to optimize the crane movement, so that the combined waiting time of the converters is minimized. For that they used Microsoft Bonsai-based reinforcement learning method.
As a result of deploying the trained RL brain to make the decision of crane assignment, there is already an improvement of 8% in daily throughput of the plant.
