"""Use the EPA crosswalk to connect EPA units to EIA generators and other data.
A major use case for this dataset is to identify subplants within plant_ids,
which are the smallest coherent units for aggregation.
Despite the name, plant_id refers to a legal entity that often contains
multiple distinct power plants, even of different technology or fuel types.
EPA CEMS data combines information from several parts of a power plant:
* emissions from smokestacks
* fuel use from combustors
* electricty production from generators
But smokestacks, combustors, and generators can be connected in
complex, many-to-many relationships. This complexity makes attribution difficult for,
as an example, allocating pollution to energy producers.
Furthermore, heterogeneity within plant_ids make aggregation
to the parent entity difficult or inappropriate.
But by analyzing the relationships between combustors and generators,
as provided in the EPA/EIA crosswalk, we can identify distinct power plants.
These are the smallest coherent units of aggregation.
In graph analysis terminology, the crosswalk is a list of edges between
nodes (combustors and generators) in a bipartite graph. The networkx python
package provides functions to analyze this edge list and extract
disjoint subgraphs (groups of combustors and generators that are connected to each other).
These are the distinct power plants. To avoid a name collision
with plant_id, we term these collections 'subplants', and identify them with a subplant_id
that is unique within each plant_id. Subplants are thus identified with the composite key
(plant_id, subplant_id).
Through this analysis, we found that 56% of plant_ids contain multiple distinct subplants,
and 11% contain subplants with different technology types, such as
a gas boiler and gas turbine (not in a combined cycle).
Usage Example:
epacems = pudl.output.epacems.epacems(states=['ID']) # small subset for quick test
epa_crosswalk_df = pudl.output.epacems.epa_crosswalk()
filtered_crosswalk = filter_crosswalk(epa_crosswalk_df, epacems)
crosswalk_with_subplant_ids = make_subplant_ids(filtered_crosswalk)
"""
from typing import Union
import dask.dataframe as dd
import networkx as nx
import pandas as pd
[docs]def _get_unique_keys(epacems: Union[pd.DataFrame, dd.DataFrame]) -> pd.DataFrame:
"""Get unique unit IDs from CEMS data.
Args:
epacems (Union[pd.DataFrame, dd.DataFrame]): epacems dataset from pudl.output.epacems.epacems
Returns:
pd.DataFrame: unique keys from the epacems dataset
"""
# The purpose of this function is mostly to resolve the
# ambiguity between dask and pandas dataframes
ids = epacems[["plant_id_eia", "unitid", "unit_id_epa"]].drop_duplicates()
if isinstance(epacems, dd.DataFrame):
ids = ids.compute()
return ids
[docs]def filter_crosswalk_by_epacems(crosswalk: pd.DataFrame, epacems: Union[pd.DataFrame, dd.DataFrame]) -> pd.DataFrame:
"""Inner join unique CEMS units with the EPA crosswalk.
This is essentially an empirical filter on EPA units. Instead of filtering by construction/retirement dates
in the crosswalk (thus assuming they are accurate), use the presence/absence of CEMS data to filter the units.
Args:
crosswalk (pd.DataFrame): the EPA crosswalk, as from pudl.output.epacems.epa_crosswalk()
unique_epacems_ids (pd.DataFrame): unique ids from _get_unique_keys
Returns:
pd.DataFrame: the inner join of the EPA crosswalk and unique epacems units. Adds the global ID column unit_id_epa.
"""
unique_epacems_ids = _get_unique_keys(epacems)
key_map = unique_epacems_ids.merge(
crosswalk,
left_on=["plant_id_eia", "unitid"],
right_on=["CAMD_PLANT_ID", "CAMD_UNIT_ID"],
how="inner",
)
return key_map
[docs]def filter_out_unmatched(crosswalk: pd.DataFrame) -> pd.DataFrame:
"""Remove unmatched or excluded (non-exporting) units.
Unmatched rows are limitations of the completeness of the EPA crosswalk itself, not of PUDL.
Args:
crosswalk (pd.DataFrame): the EPA crosswalk, as from pudl.output.epacems.epa_crosswalk()
Returns:
pd.DataFrame: the EPA crosswalk with unmatched units removed
"""
bad = crosswalk["MATCH_TYPE_GEN"].isin({"CAMD Unmatched", "Manual CAMD Excluded"})
return crosswalk.loc[~bad].copy()
[docs]def filter_out_boiler_rows(crosswalk: pd.DataFrame) -> pd.DataFrame:
"""Remove rows that represent graph edges between generators and boilers.
Args:
crosswalk (pd.DataFrame): the EPA crosswalk, as from pudl.output.epacems.epa_crosswalk()
Returns:
pd.DataFrame: the EPA crosswalk with boiler rows (many/one-to-many) removed
"""
crosswalk = crosswalk.drop_duplicates(
subset=["CAMD_PLANT_ID", "CAMD_UNIT_ID", "EIA_GENERATOR_ID"])
return crosswalk
[docs]def _prep_for_networkx(crosswalk: pd.DataFrame) -> pd.DataFrame:
"""Make surrogate keys for combustors and generators.
Args:
crosswalk (pd.DataFrame): EPA crosswalk, as from pudl.output.epacems.epa_crosswalk()
Returns:
pd.DataFrame: copy of EPA crosswalk with new surrogate ID columns 'combustor_id' and 'generator_id'
"""
prepped = crosswalk.copy()
# networkx can't handle composite keys, so make surrogates
prepped["combustor_id"] = prepped.groupby(
by=["CAMD_PLANT_ID", "CAMD_UNIT_ID"]).ngroup()
# node IDs can't overlap so add (max + 1)
prepped["generator_id"] = (
prepped.groupby(by=["CAMD_PLANT_ID", "EIA_GENERATOR_ID"]).ngroup()
+ prepped["combustor_id"].max()
+ 1
)
return prepped
[docs]def _subplant_ids_from_prepped_crosswalk(prepped: pd.DataFrame) -> pd.DataFrame:
"""Use networkx graph analysis to create global subplant IDs from a preprocessed crosswalk edge list.
Args:
prepped (pd.DataFrame): an EPA crosswalk that has passed through _prep_for_networkx()
Returns:
pd.DataFrame: copy of EPA crosswalk plus new column 'global_subplant_id'
"""
graph = nx.from_pandas_edgelist(
prepped,
source="combustor_id",
target="generator_id",
edge_attr=True,
)
for i, node_set in enumerate(nx.connected_components(graph)):
subgraph = graph.subgraph(node_set)
assert nx.algorithms.bipartite.is_bipartite(
subgraph
), f"non-bipartite: i={i}, node_set={node_set}"
nx.set_edge_attributes(subgraph, name="global_subplant_id", values=i)
return nx.to_pandas_edgelist(graph)
[docs]def _convert_global_id_to_composite_id(crosswalk_with_ids: pd.DataFrame) -> pd.DataFrame:
"""Convert global_subplant_id to an equivalent composite key (CAMD_PLANT_ID, subplant_id).
The composite key will be much more stable (though not fully stable!) in time.
The global ID changes if ANY unit or generator changes, whereas the
compound key only changes if units/generators change within that specific plant.
A global ID could also tempt users into using it as a crutch, even though it isn't stable.
A compound key should discourage that behavior.
Args:
crosswalk_with_ids (pd.DataFrame): crosswalk with global_subplant_id, as from _subplant_ids_from_prepped_crosswalk()
Raises:
ValueError: if crosswalk_with_ids has a MultiIndex
Returns:
pd.DataFrame: copy of crosswalk_with_ids with an added column: 'subplant_id'
"""
if isinstance(crosswalk_with_ids.index, pd.MultiIndex):
raise ValueError(
f"Input crosswalk must have single level index. Given levels: {crosswalk_with_ids.index.names}")
reindexed = crosswalk_with_ids.reset_index() # copy
idx_name = crosswalk_with_ids.index.name
if idx_name is None:
# Indices with no name (None) are set to a pandas default name ('index'), which
# could (though probably won't) change.
idx_col = reindexed.columns.symmetric_difference(
crosswalk_with_ids.columns)[0] # get index name
else:
idx_col = idx_name
composite_key: pd.Series = (
reindexed
.groupby('CAMD_PLANT_ID', as_index=False)
.apply(lambda x: x.groupby('global_subplant_id').ngroup())
)
# Recombine. Could use index join but I chose to reindex, sort and assign.
# Errors like mismatched length will raise exceptions, which is good.
# drop the outer group, leave the reindexed row index
composite_key.reset_index(level=0, drop=True, inplace=True)
composite_key.sort_index(inplace=True) # put back in same order as reindexed
reindexed['subplant_id'] = composite_key
# restore original index
reindexed.set_index(idx_col, inplace=True) # restore values
reindexed.index.rename(idx_name, inplace=True) # restore original name
return reindexed
[docs]def filter_crosswalk(crosswalk: pd.DataFrame, epacems: Union[pd.DataFrame, dd.DataFrame]) -> pd.DataFrame:
"""Remove crosswalk rows that do not correspond to an EIA facility or are duplicated due to many-to-many boiler relationships.
Args:
crosswalk (pd.DataFrame): The EPA/EIA crosswalk, as from pudl.output.epacems.epa_crosswalk()
epacems (Union[pd.DataFrame, dd.DataFrame]): Emissions data. Must contain columns named ["plant_id_eia", "unitid", "unit_id_epa"]
Returns:
pd.DataFrame: A filtered copy of EPA crosswalk
"""
filtered_crosswalk = filter_out_unmatched(crosswalk)
filtered_crosswalk = filter_out_boiler_rows(filtered_crosswalk)
key_map = filter_crosswalk_by_epacems(
filtered_crosswalk, epacems)
return key_map
[docs]def make_subplant_ids(crosswalk: pd.DataFrame) -> pd.DataFrame:
"""Identify sub-plants in the EPA/EIA crosswalk graph. Any row filtering should be done before this step.
Usage Example:
epacems = pudl.output.epacems.epacems(states=['ID']) # small subset for quick test
epa_crosswalk_df = pudl.output.epacems.epa_crosswalk()
filtered_crosswalk = filter_crosswalk(epa_crosswalk_df, epacems)
crosswalk_with_subplant_ids = make_subplant_ids(filtered_crosswalk)
Args:
crosswalk (pd.DataFrame): The EPA/EIA crosswalk, as from pudl.output.epacems.epa_crosswalk()
Returns:
pd.DataFrame: An edge list connecting EPA units to EIA generators, with connected pieces issued a subplant_id
"""
edge_list = _prep_for_networkx(crosswalk)
edge_list = _subplant_ids_from_prepped_crosswalk(edge_list)
edge_list = _convert_global_id_to_composite_id(edge_list)
column_order = ["subplant_id"] + list(crosswalk.columns)
return edge_list[column_order] # reorder and drop global_subplant_id