PowerIO

PowerIO is compiler infrastructure for power systems. Case files from a dozen
transmission and distribution formats parse into typed intermediate
representations (IR). Once parsed, you can perform explicit, recorded operations, like normalization, validation, and lowering.
You can compile/write the case back into any supported target format, sparse matrix families, and ML model formats.

The .pio.json package serves as a unified network payload under declared schema versions,
which records where the data came from, and how it maps back to the original source file.
Furthermore, the package contains structured diagnostics, validation, and replayable operating points, enabling many downstream tasks.

Data fidelity and interoperability is the primary goal of the PowerIO project. Writing a parsed file back to its own format returns
the original text when the reader kept it. Converting to another format writes
the modeled electrical data and reports every field the target cannot carry in
Conversion::warnings.

The core of PowerIO is written in Rust. The Rust version is used to create the Python package and the command line interface, both of which sit in this repo. The Rust implementation also enables the creation of a C ABI, which exposes PowerIO capabilities to C, C++, Julia, and other foreign function interfaces (FFIs).

Overview

PowerIO is a community infrastructure project intended to serve all developers working on electric power systems. Everyone is welcome to use, build upon, and contribute to the PowerIO infrastructure project.

Formats

Supported formats:

• MATPOWER .m
• PSS/E .raw revisions 33, 34, and 35
• PowerWorld .aux, plus read only .pwb binary cases; .pwd display files parse through the separate display API. Behavior and limits are in the format fidelity guide.
• GE PSLF .epc power flow cases
• PowerModels.jl network data JSON
• egret ModelData JSON
• pandapower pandapowerNet JSON
• PyPSA static CSV folders
• ARPA-E GO Competition Challenge 3 JSON input data
• surge .surge.json
• GridFM .parquet
• PowerIO JSON snapshots (powerio-json) and .pio.json compiler packages

Distribution networks are supported in wire coordinates via powerio-dist:

• OpenDSS .dss
• PowerModelsDistribution.jl ENGINEERING data JSON
• The (draft) BMOPF JSON spec and schema of the IEEE BMOPF task force .json

Other formats are planned; see the GitHub issues. If a format you need is missing, open an issue or a pull request. All are welcome to contribute to this community project.

Packages

This repository contains multiple packages.

powerio          # parser, Network model, source retaining writers, converters
powerio-matrix   # sparse matrices, DC sensitivity factors, graph representations
powerio-dist     # multiconductor distribution model, dss/PMD/BMOPF converters
powerio-pkg      # .pio.json compiler package envelope
powerio-cli      # the `powerio` command and ratatui TUI
powerio-py       # PyO3 extension for the Python `powerio` package
powerio-capi     # C ABI for C, C++, Julia, and other foreign function interfaces
PowerIO.jl       # Julia bindings over the C ABI

The core powerio Rust crate keeps parsing
and conversion separate from matrix, TUI, and data frame dependencies. The
Python package imports with no required
third party packages; matrix and graph helpers live behind extras.

Docs site: https://powerio.dev.
Language API map: languages guide.

Install

cargo add powerio
cargo add powerio-matrix
cargo install powerio-cli

pip install powerio
pip install 'powerio[all]'     # scipy, numpy, networkx, polars extras
pip install 'powerio[gridfm]'  # polars for Parquet inspection
pip install 'powerio[pandas]'  # pandas, pyarrow compatibility reads (Python 3.10+)

julia -e 'using Pkg; Pkg.add(url="https://github.com/eigenergy/PowerIO.jl")'

Use

Rust

use powerio::{TargetFormat, parse_file};

let parsed = parse_file("case14.m", None)?;
let net = parsed.network;
let conv = net.to_format(TargetFormat::PowerModelsJson)?;

for warning in &conv.warnings {
    eprintln!("conversion warning: {warning}");
}

std::fs::write("case14.json", conv.text)?;

Python

import powerio as pio

case = pio.parse_file("case9.m")
bprime = case.bprime()            # scipy.sparse, needs powerio[matrix]
display = pio.parse_display_file("case.pwd")
raw, warnings = pio.convert_file("case9.m", "psse")

Julia

using PowerIO

case = parse_file("case9.m")
text = to_matpower(case)
json, warnings = to_format(case, "powermodels-json")

Command line interface (CLI)

powerio convert tests/data/case14.m --to psse35 -o case14.raw
powerio convert tests/data/case14.m --to pandapower-json -o case14.pp.json
powerio convert tests/data/case14.m --to pypsa-csv -o pypsa_case
powerio convert pypsa_case --from pypsa-csv --to matpower -o case14.m
powerio convert case.epc --from pslf --to matpower -o case.m
powerio convert case.surge.json --from surge-json --to matpower -o case.m
powerio convert goc3_case.json --from goc3-json --to matpower -o case.m
powerio package tests/data/case14.m -o case14.pio.json
powerio package goc3_case.json --from goc3-json -o goc3_case.pio.json
powerio verify tests/data/case30.m --kind bdoubleprime
powerio dcopf tests/data/case30.m -o out
powerio sensitivities tests/data/case30.m -o out
powerio gridfm tests/data/case14.m -o out
powerio

Features

Current Format Fidelity

Every network reader lowers to Network. The table separates writing back to
the original file type from converting to a different file type.

PowerWorld .pwd is display data, not a network case, so it is outside this
conversion table and uses parse_display_file / parse_display_bytes. The
decoded vintages and per field evidence are maintainer notes at
powerio/src/format/powerworld/FORMAT.md.

The distribution matrix (dss, PMD JSON, BMOPF JSON, per fixture) is generated
under powerio-dist/docs/. Vendored test data keeps its own licenses next to
the fixtures under tests/data/dist/.

Known limits for every format are documented in the
format fidelity guide.

Matrices

The powerio-matrix Rust crate derives an IndexedNetwork with dense bus indices. It enables you to build common power system matrices with minimal dependencies:

• B' and B'' DCPF and FDPF matrices
• Nodal admittance matrix
• LACPF block matrix
• Signed incidence, weighted Laplacian, and flow map matrices
• PTDF and LODF sensitivity matrices
• Adjacency matrix and petgraph graph output
• Matrix Market bundles for OPF solvers
• KKT operators for OPF solvers (experimental)

Current conventions for signs, taps, phase shifts, per unit scaling, reference buses, and line parameters are documented in the matrices guide.

Normalized Form

Network::to_normalized derives a post processed copy of a case for solvers:

• powers are in per unit,
• voltage phase angles are in radians,
• inactive elements are removed,
• tap == 0 replaced with 1,
• surviving buses keep their source bus ids, and
• bus types are made consistent with generator placement and reference buses.

The normalized copy carries no retained source text, so writing it emits the derived model rather than the original file.

Python exposes the normalized form as case.to_normalized(), the C ABI as pio_normalize,
and Julia as to_normalized(case).

C ABI

powerio-capi exposes parse, query, conversion, JSON transport, normalization,
.pio.json package handles, and numeric table extraction through pio_*
functions. The public header is
powerio-capi/include/powerio.h.
Build with --features arrow to enable pio_to_arrow over the
Arrow C Data Interface.

PowerAgent

PowerIO is part of the PowerAgent community. The Python package includes an optional MCP server with tools for conversion, saving, summaries, parsing, normalization, matrix outputs, and display data.

pip install 'powerio[mcp]'
powerio-mcp

MCP clients can keep a case in the .pio.json package transport:

parsed = parse(path="case9.m", transport="package")
pkg = parsed["package_json"]
summary(package_json=pkg)
matrix("bprime", package_json=pkg)
save(out_path="case9.raw", to_format="psse", package_json=pkg)
diagnostics(pkg)

The PowerMCP bundle in PowerMCP uses the same PowerIO tool surface alongside simulator servers and bridge tools.

Compiler Packages

.pio.json packages wrap one balanced or multiconductor payload with provenance,
source maps, diagnostics, validation, summaries, lowering history, optional
derived metadata, and optional operating_points. A GO Challenge 3 package
stores the static first interval in model and the full replayable time series
in operating_points; materializing one point returns a static package with the
updates applied and the series cleared.

Rust uses powerio_pkg::NetworkPackage, Python uses the powerio.Package
class, the C ABI uses pio_package_*, and the CLI writes packages with
powerio package.

GridFM (experimental)

PowerIO writes datasets for the LF Energy open Grid Foundation Model (GridFM) project. In the command line:

powerio gridfm <case> -o <dir>

This writes the Parquet tables gridfm-datakit and
gridfm-graphkit consume under <dir>/<case>/raw/; several compatible cases
stack by scenario id.

The gridfm feature also supports reading a .parquet dataset back into a Network (read_gridfm_dataset in powerio-matrix, pio.read_gridfm in
Python), so a perturbed training scenario or a GNN predicted state can be extracted and converted back
out in any classical format:

powerio convert out/case14/raw --from gridfm --to matpower -o case14.m

The --from gridfm read path is lossy. What it recovers, what it drops, and its warning behavior
are in the format fidelity guide.

Validation

The Rust test suite covers parsers, writers, format conversion, matrix
builders, and normalization; the C ABI crate carries its own tests, and
pytest covers the Python bindings. The benchmark validation suite compares
selected outputs against PowerModels.jl, egret, ExaPowerIO.jl, and pandapower,
and imports PowerIO's PyPSA CSV folders with PyPSA. Install the oracle stack
from benchmarks/requirements.txt into the same Python 3.11+ venv that holds
the local powerio wheel.

cargo fmt --all --check
cargo test
cargo test -p powerio-capi
cargo clippy --all-targets
pytest python/tests
bash benchmarks/run_validation.sh

Benchmark method, environment, and current tables are documented in the
performance guide.

License

PowerIO is distributed under either of:

• Apache License, Version 2.0
• MIT license