novomodelo

Python bindings for the Novomodelo power systems solver.

Novomodelo is a high-performance SDDP (Stochastic Dual Dynamic Programming) solver for hydrothermal dispatch, written in Rust. This package provides Python access to case loading, validation, training, simulation, and result inspection.

Installation

pip install novomodelo-python

Prebuilt wheels are published for:

• Linux x86_64 (manylinux_2_34)
• Linux aarch64 (manylinux_2_28)
• Linux musl x86_64 (musllinux_1_2)
• macOS Apple Silicon (aarch64) and Intel (x86_64)
• Windows x86_64
• Python 3.12+

The x86_64 Linux wheel is tagged manylinux_2_34, so it needs glibc 2.34 or newer: RHEL and Rocky Linux 9+, Ubuntu 22.04+, Debian 12+. RHEL 8 and Ubuntu 20.04 are older, so pip does not select the wheel there. The aarch64 Linux wheel is tagged manylinux_2_28 (glibc 2.28 or newer).

Where no wheel fits, pip builds the published sdist instead, which needs a Rust toolchain (1.88+), cmake and a C++ compiler. The same toolchain builds the package from the root of a clone made with git clone --recurse-submodules (the solvers are vendored as submodules):

pip install ./crates/novomodelo-python

Quick Start

import novomodelo

# Load and validate a case
system = novomodelo.io.load_case("path/to/case")
print(
    f"System: {system.n_buses} buses, {system.n_hydros} hydros, {system.n_thermals} thermals"
)

# Run training + simulation
result = novomodelo.run.run("path/to/case", output_dir="output/")
print(f"Converged: {result['converged']}, LB: {result['lower_bound']:.2f}")

convergence = novomodelo.results.load_convergence("output/")
print(f"Iterations: {len(convergence)}")

simulation = novomodelo.results.load_simulation("output/")
print(f"Cost records: {len(simulation['costs'])}")

policy = novomodelo.results.load_policy("output/")
print(f"Iterations completed: {policy['metadata']['producer']['completed_iterations']}")

The same workflow, driven step by step:

# The Study lifecycle — construct, train, simulate
study = novomodelo.Study("path/to/case", output_dir="output/")
policy = study.train()
summary = study.simulate(policy)
print(f"Scenarios: {summary['n_scenarios']}, Completed: {summary['completed']}")

# Inspect the study before running any phase
print(f"Stages: {study.stochastic['n_stages']}, Seed: {study.stochastic['seed']}")

Modules

• novomodelo.io — Load and validate case directories
• novomodelo.model — Data model classes (System, Bus, Line, Thermal, Hydro, etc.)
• novomodelo.run — Execute SDDP training and simulation
• novomodelo.results — Load and inspect output artifacts, including convergence history, Parquet simulation outputs, and FlatBuffers policy (FCF) checkpoints
• novomodelo.schema — JSON Schema export for case-directory input types
• novomodelo.errors — Typed exception hierarchy for case-loading and solver errors

The package also exports Study, Policy, write_policy_checkpoint, and version_info at the top level.

Requirements

• Python >= 3.12
• No runtime dependencies (the Rust solver is statically linked)

License

Apache-2.0 — see LICENSE.

Links

• Repository
• Documentation
• Bug Tracker