Power flow solver

In mode: "pf" the power flow of every perturbed network is solved by the engine chosen with settings.pf_solver. The OPF that gives the generator set points is always solved by PowerModels.jl, whatever the value of this parameter.

settings:
  pf_solver: "lightsim2grid" # options: powermodel (default), powsybl, lightsim2grid

pf_solver	Engine	Extra to install
powermodel (default)	PowerModels.jl, through Julia. pf_fast and dcpf_fast choose between the direct and the optimizer-based solvers	none
powsybl	pypowsybl (Open Load Flow). Needs the network to be loaded with network.reader: powsybl	pip install 'gridfm-datakit[powsybl]'
lightsim2grid	lightsim2grid (C++, no Julia for the power flow)	pip install 'gridfm-datakit[lightsim2grid]'

lightsim2grid

The network is read with the native reader, and only the power flow solver changes:

network:
  source: "pglib"
  name: "case118_ieee"

settings:
  mode: "pf"
  pf_solver: "lightsim2grid"
  include_dc_res: true # the DC power flow is solved by lightsim2grid too

How the network is converted

The lightsim2grid model is built straight from the bus, generator and branch arrays of the Network, so it is the network that the perturbations have modified (loads, admittances, generator set points, outages). Buses and generators keep their order, and each branch becomes a powerline or a transformer (a branch with a non zero tap or phase shift is a transformer).

The model is not rebuilt for every perturbation: what changed since the previous power flow (branch impedances and statuses, generator statuses and set points, loads) is pushed into the same model, which lets lightsim2grid keep its solver state. It is rebuilt only if something that cannot be pushed changed (the network topology, a tap ratio or a phase shift, the shunts, the bus types, which buses have a load, or the slack generators). The model is kept by each worker process, so it is also reused from one scenario to the next.

Versions

pf_solver: lightsim2grid needs a lightsim2grid providing lightsim2grid.network.init_from_matpower. Pushing changes into the model needs the update_powerlines_parameters and update_trafos_parameters methods; without them the model is rebuilt for every power flow, which gives the same results more slowly. Until a lightsim2grid release has them, they are available by installing lightsim2grid from source.

Differences with PowerModels

• AC: the results agree with those of the fast PowerModels power flow (pf_fast: true). On case14 the difference is below 1e-6 on all the bus, generator and branch quantities.
• DC: the two solvers do not use the same DC model. lightsim2grid uses the MATPOWER one (susceptance 1 / (x * tap)) whereas PowerModels uses x / (r^2 + x^2), so the DC angles and flows differ, more on networks with large resistances.
• Slack: the active power imbalance is shared equally by the in-service generators connected to the reference bus.
• No optimizer: there is no equivalent of pf_fast: false, and pf_fast / dcpf_fast are ignored.