Optimization with PandaModels.jl


PandaModels.jl (pandapower + PowerModels.jl) is an interface (Julia package) enabling the connection of pandapower and PowerModels in a stable and functional way. Except for calling the implemented optimization models in PowerModels, users can create custom optimization models with PandaModels. Presently, users can solve some reactive power optimization problems with PandaModels.


If you are not yet using Julia, install it. For the interface to work, note that you need a version that is supported by PowerModels, PyCall and pyjulia. Currently, Former julia versions are available here.


You don’t necessarily need a Julia IDE if you are using PandaModels through pandapower, but it might help for debugging to install an IDE such as Juno. Also, PyCharm has a Julia Plugin.

  1. Add the Julia binary folder (e.g. C:UsersusernameAppDataLocalProgramsJulia-1.8.0bin` on Windows or `/Applications/Julia-1.5.app/Contents/Resources/julia/bin on MacOS) to the system variable PATH. Providing the path is correct, you can now enter the julia prompt by executing julia in your shell (on Windows, rebooting the system is needed to take advantage of changes to the PATH.

  2. The library PyCall allows to use Python from inside julia. By default, PyCall uses the Conda.jl package to install a Miniconda distribution private to Julia. To use an already installed Python distribution (e.g. Anaconda), set the PYTHON environment variable inside the Julia prompt.

    On Windows:


    On MacOS:


  3. Access the package manager by typing ]. Now install the packages: add Ipopt PowerModels PyCall. To pass the python environment variable, running build PyCall inside the julia package manager may be neccessary.

  4. Inside package manager, test your PowerModels installation by executing test PowerModels. Alternatively, you can call using Pkg and then Pkg.test("PowerModels") outside the package manager directly as julia expression. Then, test wether calling Python from Julia works, as described here.


If you cannot plot using PyCall and PyPlot in Julia, see the workarounds offered here.

  1. To call Julia from Python, install the pyjulia package with pip install julia. Afterwards, test if everything works by importing PowerModels from Python with: from julia.PowerModels import run_ac_opf. This takes some time, since Python starts a julia instance in the background, but it if the import completes without error everything is configured correctly and you can now use PowerModels to optimize pandapower networks.

  2. Additional packages are required to use the pandapower - PowerModels.jl interface with all features like TNEP or OTS. Install the “JSON” and “JuMP” packages with, e.g., julia -e 'import Pkg; Pkg.add("JSON"); Pkg.add("JuMP"); and maybe also julia -e ‘import Pkg; Pkg.add(“Cbc”); Pkg.add(“Juniper”)’ to get the TNEP and OTS libraries. Alternatively, install these packages by entering ] inside the julia console and calling add JSON add JuMP

  3. Now install our interface PandaModels.jl by type add PandaModels inside Julia package manager.

  4. Finally, you can then test whether the PandaModels.jl interface works: Navigate to your local pandapower test folder pandapower/pandapower/test/opf folder and run python-jl test_pandamodels_runpm.py or pytest test_pandamodels_runpm.py if pytest is intalled. If everything works there should be no error.

Additional Solvers

Optional additional solvers, such as Gurobi are compatible to PowerModels.jl. To use these solvers, you first have to install the solver itself on your system and then the julia interface. Gurobi is very fast for linear problems such as the DC model and free for academic usage. Let’s do this step by step for Gurobi:

  1. Download and install from Gurobi download (you’ll need an account for this)

  2. Run the file to get the gurobi folder, e.g., in linux you need to run tar -xzf gurobi<version>_linux64.tar.gz

  3. Get your Gurobi license at Gurobi license and download it (remember where you stored it).

  4. Activate the license by calling grbgetkey YOUR_KEY as described on the Gurobi license page.

  5. Add some Gurobi paths and the license path to your local PATH environment variables. In linux you can just open your .bashrc file with, e.g., nano .bashrc in your home folder and add:

# Linux

# gurobi
export GUROBI_HOME="/opt/gurobi_VERSION/linux64"
export PATH="${PATH}:${GUROBI_HOME}/bin"
# MacOS

# gurobi
export GUROBI_HOME="/Library/gurobiVERSION/mac64"
export PATH="$PATH:$GUROBI_HOME/bin"
  1. Install the julia - Gurobi interface and set the GUROBI_HOME environment with

    julia -e 'import Pkg; Pkg.add("Gurobi");'

    or type

    add Gurobi

    inside Julia package mode.

  2. Build and test your Gurobi installation by entering julia prompt and then import Pkg; Pkg.build("Gurobi"). This should compile without an error.

  3. Now, you can use Gurobi to solve your linear problems, e.g., the DC OPF, with runpm_dc_opf(net, pm_model="DCPPowerModel", pm_solver="gurobi")


The usage is explained in the PandaModels tutorial.

pandapower.runpm_ac_opf(net, pp_to_pm_callback=None, calculate_voltage_angles=True, trafo_model='t', delta=1e-08, trafo3w_losses='hv', check_connectivity=True, pm_solver='ipopt', correct_pm_network_data=True, silence=True, pm_time_limits=None, pm_log_level=0, pm_file_path=None, delete_buffer_file=True, opf_flow_lim='S', pm_tol=1e-08, pdm_dev_mode=False, **kwargs)

Runs non-linear optimal power flow from PowerModels.jl via PandaModels.jl

pandapower.runpm_dc_opf(net, pp_to_pm_callback=None, calculate_voltage_angles=True, trafo_model='t', delta=1e-08, trafo3w_losses='hv', check_connectivity=True, correct_pm_network_data=True, silence=True, pm_model='DCPPowerModel', pm_solver='ipopt', pm_time_limits=None, pm_log_level=0, delete_buffer_file=True, pm_file_path=None, pm_tol=1e-08, pdm_dev_mode=False, **kwargs)

Runs linearized optimal power flow from PowerModels.jl via PandaModels.jl

pandapower.runpm(net, julia_file=None, pp_to_pm_callback=None, calculate_voltage_angles=True, trafo_model='t', delta=1e-08, trafo3w_losses='hv', check_connectivity=True, correct_pm_network_data=True, silence=True, pm_model='ACPPowerModel', pm_solver='ipopt', pm_mip_solver='cbc', pm_nl_solver='ipopt', pm_time_limits=None, pm_log_level=0, delete_buffer_file=True, pm_file_path=None, opf_flow_lim='S', pm_tol=1e-08, pdm_dev_mode=False, **kwargs)

Runs optimal power flow from PowerModels.jl via PandaModels.jl

Flexibilities, constraints and cost parameters are defined in the pandapower element tables.

Flexibilities can be defined in net.sgen / net.gen /net.load net.sgen.controllable if a static generator is controllable. If False, the active and reactive power are assigned as in a normal power flow. If True, the following flexibilities apply:

  • net.sgen.min_p_mw / net.sgen.max_p_mw

  • net.sgen.min_q_mvar / net.sgen.max_q_mvar

  • net.load.min_p_mw / net.load.max_p_mw

  • net.load.min_q_mvar / net.load.max_q_mvar

  • net.gen.min_p_mw / net.gen.max_p_mw

  • net.gen.min_q_mvar / net.gen.max_q_mvar

  • net.ext_grid.min_p_mw / net.ext_grid.max_p_mw

  • net.ext_grid.min_q_mvar / net.ext_grid.max_q_mvar

  • net.dcline.min_q_to_mvar / net.dcline.max_q_to_mvar / net.dcline.min_q_from_mvar / net.dcline.max_q_from_mvar

Controllable loads behave just like controllable static generators. It must be stated if they are controllable. Otherwise, they are not respected as flexibilities. Dc lines are controllable per default

Network constraints can be defined for buses, lines and transformers the elements in the following columns:

  • net.bus.min_vm_pu / net.bus.max_vm_pu

  • net.line.max_loading_percent

  • net.trafo.max_loading_percent

  • net.trafo3w.max_loading_percent

How these costs are combined into a cost function depends on the cost_function parameter.


net - The pandapower format network


julia_file (str, None) - path to a custom julia optimization file

pp_to_pm_callback (function, None) - callback function to add data to the PowerModels data structure

correct_pm_network_data (bool, True) - checks if network data is correct. If not tries to correct it

silence (bool, True) - Suppresses information and warning messages output by PowerModels

pm_model (str, “ACPPowerModel”) - The PowerModels.jl model to use

pm_solver (str, “ipopt”) - The “main” power models solver

pm_mip_solver (str, “cbc”) - The mixed integer solver (when “main” solver == juniper)

pm_nl_solver (str, “ipopt”) - The nonlinear solver (when “main” solver == juniper)

pm_time_limits (Dict, None) - Time limits in seconds for power models interface. To be set as a dict like {“pm_time_limit”: 300., “pm_nl_time_limit”: 300., “pm_mip_time_limit”: 300.}

pm_log_level (int, 0) - solver log level in power models

delete_buffer_file (Bool, True) - If True, the .json file used by powermodels will be deleted after optimization.

pm_file_path (str, None) - Specifiy the filename, under which the .json file for powermodels is stored. If you want to keep the file after optimization, you should also set delete_buffer_file to False!

opf_flow_lim (str, “I”) - Quantity to limit for branch flow constraints, in line with matpower’s

“opf.flowlim” parameter:

“S” - apparent power flow (limit in MVA), “I” - current magnitude (limit in MVA at 1 p.u. voltage)

pm_tol (float, 1e-8) - default desired convergence tolerance for solver to use.

pdm_dev_mode (bool, False) - If True, the develope mode of PdM is called.

The TNEP optimization is explained in the PandaModels TNEP tutorial. Additional packages including “juniper”

pandapower.runpm_tnep(net, julia_file=None, pp_to_pm_callback=None, calculate_voltage_angles=True, trafo_model='t', delta=1e-08, trafo3w_losses='hv', check_connectivity=True, pm_model='ACPPowerModel', pm_solver='juniper', correct_pm_network_data=True, silence=True, pm_nl_solver='ipopt', pm_mip_solver='cbc', pm_time_limits=None, pm_log_level=0, delete_buffer_file=True, pm_file_path=None, opf_flow_lim='S', pm_tol=1e-08, pdm_dev_mode=False, **kwargs)

Runs transmission network extension planning (tnep) optimization from PowerModels.jl via PandaModels.jl