gridpf

Power system Power Flow as a pure-Python library: Newton-Raphson / Gauss-Seidel / DC on numpy + scipy, driven by a versioned data contract.

┌──────────────────────── gridpf.PFInput (opaque, p.u.) ───────────────────────┐
│  topology         injections          setpoints           load model (СХН)   │
│  bus_type/idx     P/Q per bus          bus_v_set           polynomial P(V)   │
│  branch r/x/b     gen − load           bus_va_set          Q(V) coefficients │
└───────────────────────────────────────────┬──────────────────────────────────┘
                                            │
                          gridpf.solve(net, options, init_v=…)
                                            │
┌───────────────────────────────────────────▼──────────────────────────────────┐
│  GS warm-start → Newton-Raphson (+ optional Q-lim outer loop) → DC fallback  │
└───────────────────────────────────────────┬──────────────────────────────────┘
                                            │
                gridpf.PFResult: V (complex p.u.), S_from/S_to (MVA),
                iterations, mismatch, convergence, failure reason

Status

Pre-alpha. The public API (PFInput / PFOptions / solve / PFResult) and the data contract version are stable enough to build on; internals may still move.

Installation

pip install gridpf

From source (development):

make setup           # venv + editable install + pre-commit hooks
# or
pip install -e ".[dev,test]"

Quick start

import numpy as np
import gridpf
from gridpf import PFInput, PFOptions, solve

# A 2-bus example: slack(0) feeding a PQ load(1) of 0.5 + j0.2 p.u. over r+jx.
z = 0.02 + 0.06j
net = PFInput(
    n_bus=2, n_branch=1,
    bus_ids=np.array([10, 11]), bus_vn_kv=np.array([110.0, 110.0]),
    bus_type=np.array([2, 0], dtype=np.int8), slack_idx=0,
    branch_ids=np.array([100]),
    from_idx=np.array([0]), to_idx=np.array([1]),
    branch_r=np.array([z.real]), branch_x=np.array([z.imag]),
    branch_g=np.zeros(1), branch_b=np.zeros(1),
    branch_g_from=np.zeros(1), branch_b_from=np.zeros(1),
    branch_g_to=np.zeros(1), branch_b_to=np.zeros(1),
    tap_ratio=np.array([1.0]), phase_shift=np.zeros(1),
    bus_g_shunt=np.zeros(2), bus_b_shunt=np.zeros(2),
    bus_p_injection=np.array([0.0, -0.5]),
    bus_q_injection=np.array([0.0, -0.2]),
)

res = solve(net, PFOptions(method="gs+nr"))
print(res.converged, np.abs(res.V))     # True [1.    0.96...]

Warm-start a re-solve from a previous result:

res2 = solve(net, PFOptions(), init_v=res.V)   # fewer NR iterations

Data contract

The input boundary is a versioned .npz file — a flat set of per-unit numpy arrays plus the contract version. Reconstruction needs only numpy, never the original model, XML, or any vendor code:

from gridpf import save_pf_input, load_pf_input_npz
save_pf_input(net, "case.npz")
net2 = load_pf_input_npz("case.npz")

The schema and its SemVer are owned by gridpf.contract (PFInput, CONTRACT_VERSION, validate_input).

Public API

Development

make check        # ruff format + lint + mypy + pytest
make test         # pytest
make type-check   # mypy gridpf
make build        # sdist + wheel

License

MIT — see LICENSE. The power-flow numerics are adapted from pandapower / PYPOWER (BSD 3-Clause); the full third-party notice and the list of adapted files are in the LICENSE file.