PowerDiff.jl

A Julia package for differentiable power system analysis. Compute sensitivities of power flow solutions, optimal power flow dispatch, and locational marginal prices with respect to network parameters.

Features

• Unified sensitivity API: calc_sensitivity(state, :operand, :parameter) with Sensitivity{T} return type
• DC OPF: B-theta formulation with analytical KKT sensitivities for demand, switching, cost, flow limits, and susceptances
• DC power flow: Switching and demand sensitivities via matrix perturbation theory
• AC power flow: Voltage and current sensitivities w.r.t. power injections
• AC OPF: Full sensitivity analysis (switching, demand, costs, flow limits) via implicit differentiation of KKT conditions
• LMP analysis: Locational marginal prices with energy/congestion decomposition
• Load shedding: Sensitivity of optimal load curtailment to network parameters

Installation

Requires Julia 1.10 or later.

using Pkg
Pkg.add("PowerDiff")

Quick Start

using PowerDiff

# Parse a case into a PowerIO module
net = parse_file("case14.m")
dc_net = DCNetwork(net)
d = calc_demand_vector(net)

# Solve DC OPF and compute sensitivities
prob = DCOPFProblem(dc_net, d)
solve!(prob)

dlmp_dd = calc_sensitivity(prob, :lmp, :d)   # dLMP/dd (n x n)
dpg_dsw = calc_sensitivity(prob, :pg, :sw)   # dg/dsw (k x m)

dlmp_dd.formulation  # :dcopf
dlmp_dd[2, 3]        # dLMP_2 / dd_3

See the Getting Started guide for DC/AC power flow and OPF walkthroughs.

Documentation

• Getting Started — DC PF, DC OPF, AC PF, AC OPF walkthroughs
• Sensitivity API — Operand/parameter tables, valid combinations, indexing
• Mathematical Background — B-theta formulation, KKT implicit differentiation
• Advanced Topics — Type hierarchy, caching, solver configuration
• API Reference — Full docstring reference

Input Format

PowerDiff reads files through PowerIO, and parse_file reads every transmission format the linked PowerIO library does — MATPOWER .m, PSS/E .raw, PowerWorld, PowerModels JSON, Egret JSON, pandapower, PyPSA, PSLF, gridfm, GO Challenge 3 and the rest. The format tokens are PowerIO's, so a reader PowerIO gains works here at once.

A path's format is inferred from its extension; a stream has no extension, so pass from (MATPOWER is assumed otherwise). A bare json names a container rather than a reader, so name the one you mean: from=:powermodels, :egret, :pandapower.

parse_file returns a PowerIO.PioModule. Beyond the case itself it carries m.diagnostics, the reader's findings as records you can branch on by code and severity; m.sources[1].format, the reader that ran; and enough for PowerIO.emit(m, "psse", path) to write the case out again.

Contributing

See CONTRIBUTING.md for how to run the tests and build the docs. Maintainers cutting a release follow RELEASING.md.

Dependencies

• PowerIO.jl — Parser and data layer (see PowerIO Integration)
• JuMP.jl — Optimization modeling
• ExaModels.jl — Alternative optimization modeling for GPU parallelization
• Ipopt.jl — Default solver for DC and AC OPF

License

Apache License 2.0