pub struct Network {
pub name: String,
pub base_mva: f64,
pub buses: Vec<Bus>,
pub loads: Vec<Load>,
pub shunts: Vec<Shunt>,
pub branches: Vec<Branch>,
pub generators: Vec<Generator>,
pub storage: Vec<Storage>,
pub hvdc: Vec<Hvdc>,
pub source_format: SourceFormat,
pub source: Option<Arc<String>>,
}Expand description
A format-neutral power network.
Fields§
§name: String§base_mva: f64§buses: Vec<Bus>§loads: Vec<Load>§shunts: Vec<Shunt>§branches: Vec<Branch>§generators: Vec<Generator>§storage: Vec<Storage>§hvdc: Vec<Hvdc>§source_format: SourceFormat§source: Option<Arc<String>>Raw source text, when read from a textual format; enables a byte-exact
same-format round-trip. Arc<String> (not Arc<str>) is deliberate: a
reader that already owns the buffer (the MATPOWER file path) moves it in
with no second copy of the whole file. The trade is one extra indirection
per access; don’t “simplify” it back to Arc<str>, which would reintroduce
the copy this avoids.
Skipped in JSON: the structured tables are the transport, not the raw
echo, and skipping also keeps serde’s rc feature out of the build. A
from_json round-trip returns this as None.
Implementations§
Source§impl Network
impl Network
Sourcepub fn in_memory(
name: impl Into<String>,
base_mva: f64,
buses: Vec<Bus>,
branches: Vec<Branch>,
) -> Network
pub fn in_memory( name: impl Into<String>, base_mva: f64, buses: Vec<Bus>, branches: Vec<Branch>, ) -> Network
A network assembled in memory from buses and branches, with no loads,
shunts, generators, storage, HVDC, or retained source document. Synthetic
topology generators and tests use it instead of repeating the struct
literal. The caller owns reference integrity (run check_references if
the ids might be inconsistent).
Sourcepub fn to_json(&self) -> Result<String, Error>
pub fn to_json(&self) -> Result<String, Error>
Serialize the structured tables to JSON — the transport the C ABI
(pio_to_json) and the Julia bridge consume. The retained source text
is excluded (see the field’s #[serde(skip)]), so the byte-exact echo
stays on the same-format write path; a from_json
round-trip reproduces every field except source, which returns None.
Sourcepub fn to_format(&self, format: TargetFormat) -> Conversion
pub fn to_format(&self, format: TargetFormat) -> Conversion
Serialize this network to format, preserving the retained source text
on same-format writes and reporting any target-format fidelity warnings.
Sourcepub fn to_matpower(&self) -> String
pub fn to_matpower(&self) -> String
Serialize this network to MATPOWER .m text.
This is byte-exact when the network was parsed from MATPOWER and still carries its retained source text.
Sourcepub fn from_json(text: &str) -> Result<Network, Error>
pub fn from_json(text: &str) -> Result<Network, Error>
Rebuild a Network from JSON produced by to_json.
Validates the result (no buses, unique bus ids, no dangling references)
before returning, so the JSON transport — the C ABI and Julia bridge ride
on it — can’t hand back a network the file readers would have rejected
(the same no-buses guard read_source applies to every parse path).
Sourcepub fn is_normalized(&self) -> bool
pub fn is_normalized(&self) -> bool
Whether this is a normalized (per-unit, radian, filtered)
derived product from to_normalized, rather
than a raw network at the file’s unit basis. Unit-sensitive code that
takes a &Network can check this instead of silently assuming MW.
Sourcepub fn check_base_mva(&self) -> Result<(), Error>
pub fn check_base_mva(&self) -> Result<(), Error>
Error unless base_mva is a positive, finite number. It is every
per-unit divisor, so a malformed base would otherwise silently poison
downstream values with NaN/Inf or flipped signs. The per-unit
consumers (to_normalized, the gridfm
export) call this; any other unit-sensitive consumer should too.
Sourcepub fn validate(&self) -> Result<(), Error>
pub fn validate(&self) -> Result<(), Error>
Check structural integrity: bus ids are unique and every element
references an existing bus. The file readers and from_json
run this; a Network built by hand (or mutated, e.g. by a scenario
generator) should call it before handing the network to
IndexedNetwork, whose dense indexing assumes it.
Source§impl Network
impl Network
Sourcepub fn to_normalized(&self) -> Result<Network, Error>
pub fn to_normalized(&self) -> Result<Network, Error>
A normalized, computation-ready copy of this network. The raw Network is
kept lossless (MATPOWER units, 1-based sparse ids, out-of-service elements
retained); to_normalized derives the form a solver or ML pipeline wants:
- Per unit (÷
base_mva): genpg/qg/pmax/pmin/qmax/qminand the ramp caps (GEN_PU_KEYS); loadp/q; shuntg/b; branchrate_a/b/c; storage energy/ratings/limits/losses; HVDCpf/pt/qf/qt, reactive limits,loss0; gen-cost coefficients (cost_to_pu). Storageps/qsand HVDC aggregatepmin/pmaxstay raw, matching the PowerModels per-unit convention. Voltages, impedances, tap, andloss1are already dimensionless. - Radians: bus
va; branchshift/angmin/angmax. - Tap:
0 → 1.0(an explicit1is kept). - Filtered: drop buses typed isolated (
BusType::Isolated) and every out-of-service element, then drop any element left referencing a dropped bus. A bus orphaned by the out-of-service filter (no in-service branch, but not typed isolated) is kept — its load is real — and surfaces as its own island, which the grounding check reports if it has no reference. - IDs: kept buses retain their source bus ids, and every surviving
endpoint stays in the same id space. Consumers that need dense rows should
use
IndexedNetwork, which derives[0, n)indices without destroying source ids. - Bus types: a bus hosting a surviving generator keeps
REFif the file marked itREF, otherwise becomesPV; a generator-less bus isPQ(so a generator-lessREFis demoted). The file’sREFbuses are kept, several included, and the consumer picks the slack. Only when no reference bus survives is the largest-pmaxin-service generator’s bus promoted toREF.
This is a derived product, not a source for write-back: source is dropped
and source_format is SourceFormat::Normalized, so writing it serializes
the per-unit/radian model instead of echoing the raw bytes, and a consumer
can tell it apart from a raw in-memory network.
Scope is the universal canonicalization only. It does not pad angle bounds,
synthesize a missing rate_a, or restrict the gen-cost model — those are
solver-prep choices a consumer applies on top. The cost rescale is
universal and lives here; the model restriction does not.
§Errors
Error::InvalidBaseMva if base_mva is not a positive, finite number
(every per-unit divisor), so a malformed base can’t silently poison the
whole network with NaN/Inf or sign-flipped values.
Error::ReferenceBusCount if no reference bus can be established — no REF
survives and there is no in-service generator to anchor one.