Package neqsim.process.ml
package neqsim.process.ml
Machine Learning and AI integration for NeqSim.
This package provides infrastructure for integrating NeqSim with modern AI/ML systems:
- Reinforcement Learning environments (Gym-compatible)
- Standardized state/action vectors for neural networks
- Constraint management for safe exploration
- Surrogate model training data export
Key Components:
StateVector- Normalized state representationActionVector- Bounded action representationConstraint- Physical/safety constraintsConstraintManager- Unified constraint handlingRLEnvironment- Base RL environment
Design Principles:
- Physics First - ML augments, never replaces, thermodynamic rigor
- Safety by Design - Constraints enforced before any action execution
- Explainability - All decisions traceable to physical constraints
- Multi-fidelity - Fast surrogates for training, full physics for deployment
- Version:
- 1.0
- Author:
- ESOL
-
ClassDescriptionStandardized action vector for RL control integration.Represents a physical or operational constraint for process equipment.Constraint category for grouping.Constraint type enumeration.Unified constraint management for process equipment.Listener interface for constraint violation events.Runner for testing RL environments with simple controllers from Java.Statistics from multiple episode runs.Result of running an episode.Adapter for extracting StateVectors from process equipment.Gymnasium (OpenAI Gym) compatible environment interface for NeqSim.Reset result matching Gymnasium API.Step result matching Gymnasium API.Reinforcement Learning environment wrapper for NeqSim process systems.Additional info from a step.Result of a simulation step.Standardized state vector for physics-grounded world models and RL integration.Interface for process equipment that can export standardized state vectors.Training data collector for surrogate model development.Feature definition for inputs/outputs.Running statistics calculator.