Skip to content

ADR-0010: The electrical network enters the optimization: linear power flow, line expansion, security constraints and transmission economics

  • Status: accepted
  • Date: 2026-10-24

Context

ADR-0008 and ADR-0009 describe the AC network in the canonical model and solve it as a power flow of a fixed dispatch. In every optimization mode the same arcs still become controllable Links: the optimizer sees capacities and efficiencies, never Kirchhoff's voltage law, and expands a Link when an arc is expandable. PyPSA's distinctive capability — dispatch and expansion under the linear power-flow equations, with the option of surviving the outage of any branch — therefore stays unreachable, although every input it needs is already in the payload. Transmission economics were a second gap: an arc carried no lifetime or interest rate, so its investment cost could not be annualized and was warned about and dropped on PyPSA.

PyPSA 1.2.4 builds the voltage-law constraints automatically when a sub-network contains Lines with a reactance, expands s_nom_extendable Lines (with the impedance held fixed, the documented approximation), honors s_max_pu as a loading limit, and solves the security-constrained problem with optimize_security_constrained(branch_outages).

Decision

The run specification gains experiment.network:

  • model: transport (the default and today's behaviour: every arc is a Link) or linear_power_flow. With the latter PyPSA emits the electrical carrier's buses with their voltage level, every arc that carries electrical as a Line or a Transformer — s_nom from the rating, s_nom_extendable, s_nom_min and s_nom_max from the arc's capacity, s_max_pu from the new electrical.max_loading, capital_cost from the arc's costs — and solves the optimization under the voltage law. Arcs without electrical stay Links, so a controlled tie in an otherwise electrical network is expressed by leaving the impedance out.
  • security.outages: a list of electrical arcs whose single outage the dispatch must survive; PyPSA runs the security-constrained solve. It needs model: linear_power_flow and mode plan, and it cannot carry the portable constraints the sidecar applies, like the operate and stochastic runners.

The network model is a capability, linear_power_flow, which PyPSA alone claims; every other target rejects model: linear_power_flow the way it rejects an unsupported mode, because a transport solve of a payload that asked for the voltage law is a different problem, not an ignorable input. The transport default keeps every existing payload and every other target unchanged.

Transmission gains lifetime, interest_rate and cost_basis, and PyPSA emits capital_cost on Links and Lines from investment_per_capacity plus investment_per_capacity_distance times distance, annualized like a technology's investment; fixed O&M folds in the same way. Calliope keeps its per-distance cost mapping and the other transmission cost fields keep their warning there; nothing it accepted changes meaning.

Every optimization run now writes a results/ folder next to network.nc: per-component dispatch and flows, storage states, bus marginal prices, the optimal capacities and PyPSA's statistics table, so the API consumer that reads the power-flow tables reads the planning tables the same way.

Consequences

  • Network-constrained dispatch and line expansion become one switch on a payload that already carries the electrical fields; the same payload runs as a transport model on Calliope. The consumer decides per call.
  • Line expansion holds the impedance fixed. The iterative reactance update (optimize_transmission_expansion_iteratively) is a possible later option; it is not claimed.
  • Security constraints multiply the problem by the number of outages; the list is explicit so the cost is visible, and all is deliberately absent.
  • Out of scope: DC sub-networks, reactive power inside the optimization (linear power flow is active-power only), and line-type catalogues.