How lollcat/rl-process-design works
This repository explores reinforcement-learning-based chemical process synthesis through two documented case studies: choosing reactor sequences and designing distillation-column trains. The principal workflow is an agent selecting process operations and operating parameters, a simulator evaluating the resulting process and reward, and training/testing components refining or assessing the policy. The sampled continuous-action runner explicitly constructs the simulator and workers; the worker’s detailed environment/model calls are not fully shown, so those links are marked unverified. The README documents the reactor case study, while the tree and source index identify the distillation and additional reactor experiments.
Main entry points: main.py Training runner · Worker_constrained_6.py RL worker · Simulator_New.py Distillation simulator · Simulator_new_reward.py Alternative reward simulator
Components 19
Reactor studies
- Reactor sequencing
Discrete/PFR or CSTR/PFRorCSTR.py - Reactor cooling
Discrete/ReactOrCool/ReactorCool.py
Distillation design
- Distillation agent
Discrete/Distillation/Distillation_simple1.py - Distillation simulator
Discrete and Continuous/Env/Simulator_New.py - Alternative reward simulator
Discrete and Continuous/Env/Simulator_new_reward.py
Learning runtime
- Training runner
Discrete and Continuous/main.py - RL worker
Discrete and Continuous/Workers/Worker_constrained_6.py - Continuous policy
Discrete and Continuous/Nets/P_actor.py - Discrete Q network
Discrete and Continuous/Nets/DQN_dueling_new_structure.py - DQN variants
Discrete and Continuous/Nets/DQN.py - Dueling DQN
Discrete and Continuous/Nets/DQN_dueling.py - Policy evaluator
Discrete and Continuous/Utils/tester.py - Action exploration
Discrete and Continuous/Utils/OrnsteinNoise.py - Experience memory
Discrete and Continuous/Utils/memory.py
Process calculations
- Column sizing
Discrete and Continuous/Env/Sizing/column_sizing.py - Capital costing
Discrete and Continuous/Env/Sizing/dc_capital_cost.py - Operating costing
Discrete and Continuous/Env/Sizing/dc_annual_cost.py - Sizing simulator
Discrete and Continuous/Env/Sizing/DistillationSimulator.py
Other components
- Process designer
Connections 27
- Process designer → Reactor sequencing: specifies sequence
README.md - Process designer → Training runner: starts training
- Reactor sequencing → Process designer: produces design
- Training runner → RL worker: creates workers
Discrete and Continuous/main.py - Training runner → Distillation simulator: configures environment
Discrete and Continuous/main.py - Training runner → Continuous policy: selects model
Discrete and Continuous/main.py - Training runner → Discrete Q network: selects model
Discrete and Continuous/main.py - Training runner → DQN variants: selects model
Discrete and Continuous/main.py - Training runner → Dueling DQN: selects model
Discrete and Continuous/main.py - Training runner → Policy evaluator: imports evaluator
Discrete and Continuous/main.py - RL worker → Distillation simulator: steps environment
- RL worker → Continuous policy: updates policy
- RL worker → Discrete Q network: updates Q network
- RL worker → Action exploration: uses exploration
Discrete and Continuous/Workers/Worker_constrained_6.py - RL worker → Experience memory: uses replay memory
- Policy evaluator → Distillation simulator: evaluates policy
Discrete and Continuous/Utils/tester.py - Distillation simulator → Column sizing: sizes columns
Discrete and Continuous/Env/Simulator_New.py - Distillation simulator → Capital costing: calculates capital cost
Discrete and Continuous/Env/Simulator_New.py - Distillation simulator → Operating costing: calculates operating cost
Discrete and Continuous/Env/Simulator_New.py - Alternative reward simulator → Column sizing: sizes columns
Discrete and Continuous/Env/Simulator_new_reward.py - Alternative reward simulator → Capital costing: calculates capital cost
Discrete and Continuous/Env/Simulator_new_reward.py - Alternative reward simulator → Operating costing: calculates operating cost
Discrete and Continuous/Env/Simulator_new_reward.py - Sizing simulator → Column sizing: uses sizing
Discrete and Continuous/Env/Sizing/DistillationSimulator.py - Sizing simulator → Capital costing: uses costing
Discrete and Continuous/Env/Sizing/DistillationSimulator.py - Sizing simulator → Operating costing: uses costing
Discrete and Continuous/Env/Sizing/DistillationSimulator.py - Process designer → Reactor cooling: explores reactor design
- Process designer → Distillation agent: explores distillation