Installation

Requirements

• Julia ≥ 1.10
• PowerModels.jl
• PowerModelsACDC.jl
• At least one solver appropriate to the formulation you intend to use (see below)

Installing the package

The package is not yet in the Julia General registry, so install it directly from GitHub:

   ] add PowerModelsTopologicalActions

Or, for development:

$ git clone https://github.com/Electa-Git/PowerModelsTopologicalActions.jl
$ cd PowerModelsTopologicalActions.jl
$ julia --project=. -e 'using Pkg; Pkg.instantiate()'

Then load it. The module name is long, so an alias is conventional:

using PowerModels;        const _PM     = PowerModels
using PowerModelsACDC;    const _PMACDC = PowerModelsACDC
using PowerModelsTopologicalActions; const _PMTP = PowerModelsTopologicalActions

Choosing solvers

Which solver you need depends entirely on the formulation, because the formulation determines the problem class. Getting this pairing wrong is the single most common source of confusing failures. I recommend using Gurobi for the solving the topology optimization models. If you do not have access to it, I made available a test of the AC-feasibility check with some results obtained with Gurobi. Using a free solver instead of Gurobi is still something which is work in progress.

A working solver stack

This is the configuration used to produce the published results:

using JuMP, Ipopt, Gurobi, Juniper

gurobi  = JuMP.optimizer_with_attributes(Gurobi.Optimizer,
              "MIPGap" => 1e-4)

ipopt   = JuMP.optimizer_with_attributes(Ipopt.Optimizer,
              "tol" => 1e-6,
              "print_level" => 0)

juniper = JuMP.optimizer_with_attributes(Juniper.Optimizer,
              "nl_solver"  => ipopt,
              "mip_solver" => gurobi,
              "time_limit" => 36000)

Use juniper for ACPPowerModel, gurobi for everything else. ipopt is more stable than gurobi for the LPACCPowerModel simulation of the OPF model. gurobi should be used for every grid topology optimization model.

Faster Ipopt

For the larger cases, Ipopt's default linear solver is a bottleneck. If you have access to the HSL library, ma97 is markedly faster:

using HSL_jll
ipopt = JuMP.optimizer_with_attributes(Ipopt.Optimizer,
            "tol" => 1e-6, "print_level" => 0, "linear_solver" => "ma97")

Solver settings for the models

Both PowerModels and PowerModelsACDC take a setting dictionary. Topological-action problems need branch flows reported in the solution, and converter losses modelled:

s = Dict("output" => Dict("branch_flows" => true), "conv_losses_mp" => true)

# if you also want dual variables from the convex/linear formulations
s_dual = Dict("output" => Dict("branch_flows" => true, "duals" => true),
              "conv_losses_mp" => true)

Pass it through as setting = s on every run_* call.

Bundled test cases

Test networks live in test/data_sources/