4.powerflow

4. Power Flow

This step runs time-series low-voltage power-flow calculations for each scenario/grid file created by the previous steps. It evaluates the grid before and after DER expansion (PV/HP/etc.) using active and reactive power demand time series, and writes the results back into a copy of the scenario .h5 file.

At a high level:

• Reads a pandapower network from the input .h5.
• Builds pre-expansion and post-expansion per-bus $P/Q$ demand time series.
• Runs pandapower power flow for each timestep (optionally in parallel).
• Stores voltages, line loadings, and external-grid imports into the output .h5.

Folder & file structure

4.powerflow/
  config.py
  run_pwrflw.py
  run_cluster_serialstd.sh
  start_batch_jobs_serialstd.sh
  environment.yml
  environment_HPC.yml
  Input/
    *.h5
  Output/
    *.h5   (copied from Input/ and augmented with pwrflw/* datasets)
  logs/
    normal/
    errors/
  src/
    demands.py
    powerflow.py
    save_grid.py
    grid_topol.py
    resource_report.py

Notes:

• Input/ and Output/ are controlled by config.py (Config.DATA_DIR, Config.STORAGE_DIR).
• run_pwrflw.py always reads from Input/ and writes to Output/.

Required inputs

1) Scenario .h5 file in Input/

The code expects each input file to contain (HDF5 keys / datasets):

• raw_data/net: a pandapower network serialized as a JSON string (loaded via pandapower.from_json_string).
• urbs_in/demand: raw household electricity demand time series (active power), used for the pre-expansion case.
• urbs_out/MILP/tau_pro: urbs optimization results, used to reconstruct post-expansion net imports, PV generation, and HP demand.

The exact schema of these tables is defined by upstream steps; this step assumes they match what src/save_grid.py and src/demands.py read.

2) Python environment

Dependencies are listed in:

• environment.yml (local/dev)
• environment_HPC.yml (cluster)

Core runtime packages used in this step:

• pandapower
• pandas, numpy, h5py, tables
• multiprocessing (standard library)

Generated outputs

For each processed input file, an output file is created in Output/ with the same filename. The output file is a copy of the input .h5 augmented with additional datasets.

Written HDF5 keys

run_pwrflw.py writes:

• /pwrflw/input/demand_pre: per-bus pre-expansion demand time series (active + reactive).

• /pwrflw/input/demand_post: per-bus post-expansion demand time series (active + reactive).

• /pwrflw/output/pre/demand_import: external grid import (p/q) per timestep.

• /pwrflw/output/pre/vm: bus voltage magnitudes vm_pu per timestep.

• /pwrflw/output/pre/line_loads: line flows and currents per timestep.

• /pwrflw/output/post/demand_import: external grid import (p/q) per timestep.

• /pwrflw/output/post/vm: bus voltage magnitudes vm_pu per timestep.

• /pwrflw/output/post/line_loads: line flows and currents per timestep.

src/demands.py additionally writes reactive-power components derived from urbs results:

• pwrflw/urbs_out/MILP/reactive: concatenated reactive components (household, heat pump, PV) for traceability.

Logs

On the HPC submission scripts, stdout/stderr are written to:

• logs/normal/<jobid>_output.log
• logs/errors/<jobid>_error.log (removed automatically if empty)

How the code works (conceptual)

Demand reconstruction (src/demands.py)

• Pre-expansion: household active power is taken from urbs_in/demand, and reactive power is synthesized using a fixed power factor (config.PF_ELC).
• Post-expansion: the urbs results urbs_out/MILP/tau_pro are filtered to keep:
  • import and feed_in (to compute net imports),
  • heatpump_air (heat pump load),
  • all rooftop PV technologies (pro starting with Rooftop...).

Reactive power post-expansion is computed using:

• HP reactive demand from config.PF_HP.
• PV reactive capability bounded by config.PF_PV_MIN. PV is assumed to produce reactive power to reduce net reactive import as much as possible within its capability.

Grid preparation (src/powerflow.py)

Before running the time-series power flow, the network is “relaxed” to avoid artificial constraint binding:

• Line current limits max_i_ka are set very high.
• Load max_p_mw limits are set very high.
• Bus voltage bounds are widened.
• The transformer is removed and replaced by a closed bus-bus switch between HV/LV sides.

Power flow execution (src/powerflow.py)

For each timestep:

• Loads are updated in grid.load (bus-indexed update).
• pandapower.runpp(..., algorithm="bfsw") is executed.
• Results are collected:
  • external-grid import from res_ext_grid
  • bus voltages from res_bus["vm_pu"]
  • line flows/currents from res_line

Optional parallelization uses Python multiprocessing by chunking timesteps across workers.

Running

Local run

1. Put one or more scenario files into Input/.

2. Run for a specific file ID prefix:

python3 run_pwrflw.py 0 --n_cpu 8

The positional argument (0 above) is matched against the prefix before the first underscore in the filename, e.g.:

• 0_N2819500E4261500_... .h5 matches inputfile_id=0.

Cluster (Slurm)

Submit a single job:

sbatch run_cluster_serialstd.sh 0

Submit a range of jobs (START..END, inclusive):

bash start_batch_jobs_serialstd.sh 0 24

run_cluster_serialstd.sh activates conda env pwrflw-hpc and passes --n_cpu $SLURM_CPUS_PER_TASK.

Details to keep in mind

• File selection logic: run_pwrflw.py chooses the first .h5 in Input/ whose prefix matches inputfile_id. If multiple files share the same prefix, only the first match will be used.
• Output overwrites: if Output/<filename> already exists it will be overwritten (because the input file is copied over at start).
• Units: loads are converted from kW/kVAr to MW/MVAr by dividing by 1000 before running pandapower.
• Parallel runs: each worker receives a deep-copied pandapower net. This avoids shared-state issues but increases memory use.
• Reactive power sign convention: the code models inductive/lagging demand as negative Q (see src/demands.py). Ensure upstream/downstream steps use consistent conventions.
• Performance: runtime scales with #timesteps × #buses/lines. Use --n_cpu to distribute timesteps.

Troubleshooting

• If you see IndexError or “no matched files”, verify that:
  • the file exists in Input/,
  • it ends with .h5,
  • its prefix before the first _ matches the inputfile_id you pass.
• If pandapower fails to converge for some timesteps, consider checking the input demands and the integrity of the network (zero-length lines etc.). Helper utilities exist in src/grid_topol.py.