Operating policy generation using a reinforcement learning agent in a melt facility
This study presents a methodology to allow a reinforcement learning agent to generate near-optimal policies for a melt facility. The application of the learning method to this industrial scale, dynamic, stochastic problem poses a number of challenges. The process is formulated as a semi-Markov Decision Problem. A novel method for application of RL agents to continuous state and action spaces, based on mapping continuous to discrete state and action spaces is developed. The agent successfully identified robust polices that improved on the best-practice of expert operators.
Interesting approach to solve this complex problem. Have you explored any other optimisation techniques that could be applied to generate operating policies for this particular problem? I'm thinking if this problem (metal transfer operation) could be formulated to be solved by Evolutionary Computation techniques such as Genetic Algorithms or Particle Swarm Optimisation. Thank You.
Gonzalo
I agree with Gonzalo: multi-agent systems are interesting for distributing intelligence, but their capability for giving near optimal solutions depends in my opinion on the problem they have to solve. I would be more confident with GA which explore many solutons, instead of the agent which makes in my opinion a local decision.
Who's online
There are currently 1 user and 134 guests online.
- phaas
What platform was used to develop the agent?