Overview¶

Flowcean is a research-oriented Python toolkit for defining, simulating, identifying, evaluating, and reproducing models of cyber-physical systems.

Flowcean originates from research into automatic model generation for CPS. Its modular architecture combines environments, transforms, learning strategies, metrics, adapters, and integrations with established model-learning tools. Hybrid dynamical systems are a first-class modeling approach within this broader toolkit, not a requirement for every Flowcean study.

Modeling Approaches¶

Flowcean supports different study designs rather than prescribing a single pipeline.

Hybrid-System Modeling¶

Use flowcean.hybrid to define locations, flows, event surfaces, transitions, resets, parameters, and initial conditions. Native simulation produces hybrid trajectories that can be sampled into trace frames, while HyDRA identifies flow models and selectors from observed data.

Data-Driven Model Learning¶

Compose learning strategies with recorded datasets, incremental streams, active environments, simulations, or connected CPS data sources. Backends such as scikit-learn, PyTorch, River, PySR, and external learners provide concrete learning algorithms.

Evaluation and Integration¶

Evaluate models with task-appropriate metrics or trajectory comparisons. Environments, transforms, learners, models, metrics, and adapters can be combined according to the needs of each study and reused across hybrid and non-hybrid applications.

Main Interfaces¶

• flowcean.hybrid provides hybrid-system definitions, simulation, trace conversion, and plotting; use flowcean.hybrid.benchmarks for reusable systems and flowcean.hybrid.hydra for identification.
• flowcean.core provides environments, learners, models, metrics, transforms, callbacks, and learning strategies.
• flowcean.polars provides dataframe environments, datasets, time-series support, and reusable transforms.
• Backend packages such as flowcean.sklearn, flowcean.river, flowcean.torch, and flowcean.pysr connect concrete learning algorithms.
• Adapters connect Flowcean studies to CPS data sources; flowcean.aalpy learns automata locally from traces.

See the modules for the established conceptual architecture, the API reference for implementation details, and the examples for runnable studies. The initial Flowcean concepts and research context are presented by Knitt et al. ¹.