Skip to content

Filtering loadflow results#

A contingency analysis produces one result row per timestep, contingency, element and side. Most of those rows say nothing: a branch loaded to 12% of its rating, a busbar sitting at nominal voltage. They are computed, carried between workers, and written to disk exactly like a violation is — and in an optimization run, a full result set is stored for every candidate topology that gets evaluated.

A LoadflowResultFilter drops the rows that carry no decision value, as they are produced. It is off by default: without one, results are exactly what they always were.

Furthermore, the used filter is stored along the results as result_filter — None when nothing was filtered — so a row that is absent by policy can be told apart from one that was never computed.

What it filters#

Branch results and node results. Every other table — convergence, voltage-angle differences, regulating elements, SpPS, cascade, connectivity — is small and passes through untouched.

How a row is judged#

  • Any active threshold can keep a row. They are alternatives, not conditions to satisfy together.
  • Unknown is never dropped. A branch with no rated current, a bus with no voltage limits, or any row from a contingency that did not converge has no value to compare, so it is kept.
  • The base case is exempt, unless you say otherwise in retain_basecase.

Using it#

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
from toop_engine_contingency_analysis.ac_loadflow_service import get_ac_loadflow_results
from toop_engine_interfaces.loadflow_result_filter import (
    BranchLoadflowResultFilter,
    LoadflowResultFilter,
    NodeLoadflowResultFilter,
)

result_filter = LoadflowResultFilter(
    # Keep branches at or above 70% of their rated current.
    branch_filters=BranchLoadflowResultFilter(loading_above=0.7),
    # Keep buses within the top 30% of their voltage band, plus any bus that moves
    # more than 5% between N-0 and N-1 even while it stays inside the band.
    node_filters=NodeLoadflowResultFilter(vm_loading_above=0.7, vm_basecase_deviation_above=5.0),
)

results = get_ac_loadflow_results(net, nminus1_definition, result_filter=result_filter)

The two backends take the same policy directly:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
# pandapower
run_contingency_analysis_pandapower(
    net, nminus1_definition, job_id, timestep,
    cfg=ContingencyAnalysisConfig(method="ac", polars=True, result_filter=result_filter),
)

# powsybl
run_contingency_analysis_powsybl(
    net, nminus1_definition, job_id, timestep, method="ac", polars=True, result_filter=result_filter,
)

To filter results you already have, rather than as they are computed:

1
2
3
from toop_engine_contingency_analysis.result_filter import apply_result_filter

filtered = apply_result_filter(results, result_filter, nminus1_definition.base_case.id)