AlphaDow: Reinforcement Learning for Industrial Production Scheduling
summary
Reinforcement learning trains agents that generate production schedules for a 5-product plant, enforcing action masking to block illegal transitions and minimum campaign lengths; initial random agents show 46% availability and very high cost, and trained agents recover sensible cycles, beat the fixed product‑wheel baseline, and lower costs. Training uses a custom Advantage Actor–Critic, then scales with Ray + Azure and Ray Tune for population‑based tuning; evaluation runs Monte Carlo year‑long demand samples and the system is now in pilot testing against real plant data.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:32Industrial Production Scheduling
- 5:30Objectives
- 7:17AlphaDow
- 8:38Proof of Concept
- 12:23Training Results
- 14:16Stochasticity
- 17:06Results
- 19:25Deployment
related talks
35mIndustrial Applications of Reinforcement Learning
22mAWARE: Automate Workload Autoscaling with Reinforcement Learning in Production...
47mHow agentic AI solves complex scheduling problems
55mAlgorithmic Trading and Machine Learning
19mDynaRL: Flexible and Dynamic Scheduling of Large-Scale Reinforcement Learning Training
12mMachine Learning for Adaptive, Self-Optimizing Systems
32mFrom Three Days to One Minute: Architecture Trade-offs in Production Enterprise AI
55mAutomate Inventory Rebalancing with Agentic AI