Skip to content

Rung 14: two futures and a risk preference — capacity chosen once, dispatch per scenario

One rung of the PyPSA corpus: the file pypsa.yaml projected onto what this network builds, attached to that network, and held to what PyPSA solves it to.

✔ Verified against pypsa master at 3625c816 — objective 9267.386667 on both sides; structure ✔ 12 constraints · 6 variables, name for name; size ✔ 87 rows · ✔ 37 columns · ✔ 148 nonzeros; duals ✔ 87 rows; objective only — the linopy oracle (tests/linopy_lane) stops at OracleCannotBuildError: the linopy lane cannot build a quadratic constraint, and no reformulation of it is exact. The language accepts it and specsolve builds it, so this is a limit of the lane rather.

Rows and columns, PyPSA against specsolve, name for name
row PyPSA specsolve
Bus-nodal_balance 16 16
CVaR-def 1 1
CVaR-excess-calm 1 1
CVaR-excess-stormy 1 1
Generator-ext-p-lower 8 8
Generator-ext-p-upper 8 8
Generator-ext-p_nom-lower 2 2
Generator-ext-p_nom-upper 2 2
Generator-fix-p-lower 16 16
Generator-fix-p-upper 16 16
Link-fix-p-lower 8 8
Link-fix-p-upper 8 8
column PyPSA specsolve
CVaR 1 1
CVaR-a 2 2
CVaR-theta 1 1
Generator-p 24 24
Generator-p_nom 1 1
Link-p 8 8

The model

The same model, as math

The file states a plain n.optimize(), its multi-period class under the multi_investment_periods keyword, and its stochastic class, which n.set_scenarios(...) sets. Every quantity chosen once the future is known (second-stage) spans the futures dispatch is chosen in (scenario). Capacity is chosen once, before the future is known, and paid at its expected cost in each investment period its build year and lifetime span. Operation is the expectation over the scenarios' weights, with a share priced at the worst-outcome mean cost (CVaR) only where a risk preference is set. A plain run's data has one scenario, one period, unit weights, no outage and every active flag true, so the problem is the standard one. On a security-constrained run, PyPSA copies each branch flow limit once per outage, and the outage factors, active flags and scenario weights come from data prep.

Sets

Symbol Meaning
\(\mathcal{B}\) index \(b\) — breakpoint — the corners of a piecewise-linear curve, in order. A curve runs over the breakpoints its data has rows for, so curves may differ in length
\(\mathcal{N}\) index \(n\) — bus with \(\mathrm{Generator\_bus}: \mathcal{G} \to \mathcal{N},\ \mathrm{Link\_bus0}: \mathcal{L} \to \mathcal{N},\ \mathrm{Link\_output\_bus}: \mathcal{O} \to \mathcal{N},\ \mathrm{Load\_bus}: \mathcal{D} \to \mathcal{N},\ \mathrm{Process\_output\_bus}: \mathcal{R} \to \mathcal{N}\) — network nodes
\(\mathcal{G}\) index \(g\) — generator with \(\mathrm{Generator\_bus}: \mathcal{G} \to \mathcal{N}\) — generating units, each on one bus
\(\mathcal{L}\) index \(l\) — link with \(\mathrm{Link\_bus0}: \mathcal{L} \to \mathcal{N},\ \mathrm{Link\_output\_link}: \mathcal{O} \to \mathcal{L}\) — controllable connections, each from one bus to the buses it delivers to
\(\mathcal{O}\) index \(o\) — link_output with \(\mathrm{Link\_output\_bus}: \mathcal{O} \to \mathcal{N},\ \mathrm{Link\_output\_link}: \mathcal{O} \to \mathcal{L}\) — a link's output ports, one label per port a link declares. The labels are PyPSA's bus1, bus2, … columns read long, data prep. A link of any number of output ports is then one term in the balance
\(\mathcal{D}\) index \(d\) — load with \(\mathrm{Load\_bus}: \mathcal{D} \to \mathcal{N}\) — demands, each on one bus
\(\mathcal{Y}\) index \(y\) — period with \(\mathrm{snapshot\_period}: \mathcal{T} \to \mathcal{Y}\) — investment periods. It is PyPSA's investment_periods
\(\mathcal{J}\) index \(j\) — process with \(\mathrm{Process\_output\_process}: \mathcal{R} \to \mathcal{J}\) — generalised multi-port converters, each with an internal power that every port draws or delivers at its own rate
\(\mathcal{R}\) index \(r\) — process_output with \(\mathrm{Process\_output\_bus}: \mathcal{R} \to \mathcal{N},\ \mathrm{Process\_output\_process}: \mathcal{R} \to \mathcal{J}\) — a process's ports, one label per port a process declares. The labels are PyPSA's bus0, bus1, …, each with a signed rate, data prep. A process of any number of ports is then one term in the balance
\(\Xi\) index \(\xi\) — scenario — the futures dispatch is chosen in, each with a weight
\(\mathcal{T}\) index \(t\) — snapshot with \(\mathrm{snapshot\_period}: \mathcal{T} \to \mathcal{Y}\) — dispatch periods

Parameters

Symbol Meaning
\(\mathrm{Bus\_attached}\) Bus_attached over \(\mathcal{N} \times \mathcal{T}\) — whether a load, a unit, a port or a branch that is active in the snapshot's period is attached to the bus, data prep. A bus nothing is attached to has no balance row
\(\alpha\) CVaR_alpha (scalar), absent where the data has no row — PyPSA's risk_preference['alpha'] — the confidence level. The tail is the other 1 - alpha of the probability. CVaR_alpha has a value exactly where CVaR_omega has one
\(\omega\) CVaR_omega (scalar), \(0\) where the data has no row — PyPSA's risk_preference['omega'] — the share of operating cost priced at the tail rather than in expectation. CVaR_omega has a value only where a risk preference is set. Without a risk preference, the file has no CVaR variables or rows. With omega = 0, PyPSA builds them, and the optimum is the risk-neutral one
\(\mathrm{on}\) Generator_active over \(\mathcal{G} \times \mathcal{T}\) — whether a generator is active in a snapshot's period. It is PyPSA's active, from build year and lifetime, data prep
\(\mathrm{act}\) Generator_active_asset over \(\mathcal{G}\) — whether a generator is one of PyPSA's active_assets, by its active flag alone, without build year and lifetime. A unit that is active in no snapshot of the horizon still counts. PyPSA refuses one that differs by scenario (constants.py:54)
\(\mathrm{c}^{\mathrm{cap}}\) Generator_capital_cost over \(\mathcal{G} \times \Xi\) — cost of one unit of nominal power for the modelled horizon (PyPSA's periodized_cost), data prep. It is overnight_cost as an annuity over lifetime at discount_rate, times nyears, where overnight_cost is given, and capital_cost where it is not. Both add fom_cost (components.py:1157-1178, costs.py:102-203)
\(\mathrm{W}\) Generator_capital_weight over \(\mathcal{G}\) — the sum of period weights a generator is active in. It is PyPSA's active * period_weighting, summed, data prep
\(\mathrm{com}\) Generator_committable over \(\mathcal{G}\) — whether output is gated by an on/off status decision
\(\mathrm{c}\) Generator_marginal_cost over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — cost of one unit of output
\(\mathrm{p}^{\mathrm{curve}}\) Generator_marginal_cost_curve_p over \(\mathcal{B} \times \mathcal{G}\), absent where the data has no row — the output at each breakpoint of a generator's piecewise marginal_cost curve: PyPSA's p_pu breakpoints times p_nom, data prep. A generator with no curve has no rows
\(\mathrm{c}^{(2)}\) Generator_marginal_cost_quadratic over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — cost of the square of one unit of output
\(\overline{\mathrm{p}}\) Generator_p_max_pu over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — most output, per unit of nominal power: an availability profile
\(\underline{\mathrm{p}}\) Generator_p_min_pu over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — least output, per unit of nominal power
\(\mathrm{p}^{\mathrm{nom}}\) Generator_p_nom over \(\mathcal{G} \times \Xi\) — nominal power
\(\mathrm{ext}\) Generator_p_nom_extendable over \(\mathcal{G}\) — whether the nominal power is a decision
\(\overline{\mathrm{p}}^{\mathrm{nom}}\) Generator_p_nom_max over \(\mathcal{G} \times \Xi\) — most nominal power an extendable generator may be built at
\(\underline{\mathrm{p}}^{\mathrm{nom}}\) Generator_p_nom_min over \(\mathcal{G} \times \Xi\) — least nominal power an extendable generator may be built at
\(\mathrm{sgn}\) Generator_sign over \(\mathcal{G}\) — the sign output enters its bus's balance with. It is PyPSA's sign, 1 unless given, -1 for a unit that draws power. PyPSA refuses one that differs by scenario (constants.py:46)
\(\mathrm{on}^{f}\) Link_active over \(\mathcal{L} \times \mathcal{T}\) — whether a link is active in a snapshot's period. It is PyPSA's active, data prep
\(\mathrm{com}^{f}\) Link_committable over \(\mathcal{L}\) — whether flow is gated by an on/off status decision
\(\eta\) Link_efficiency over \(\mathcal{O} \times \Xi \times \mathcal{T}\) — share of the flow that arrives at an output port, PyPSA's efficiency, efficiency2, … read long: negative where that port consumes rather than delivers. Read at the snapshot the flow arrives, so a delayed port delivers at its arrival snapshot's efficiency (constraints.py:1575)
\(\mathrm{c}^{f}\) Link_marginal_cost over \(\mathcal{L} \times \Xi \times \mathcal{T}\) — cost of one unit of flow
\(\mathrm{c}^{f,(2)}\) Link_marginal_cost_quadratic over \(\mathcal{L} \times \Xi \times \mathcal{T}\) — cost of the square of one unit of flow
\(\mathrm{cyc}^{f}\) Link_output_cyclic_delay over \(\mathcal{O}\) — whether a delayed port's flow wraps from the end of its investment period (PyPSA's cyclic_delay, cyclic_delay2, …). Where it does not, the flow still in transit at each period's first snapshots is lost. The flag is the same in every scenario, as the delay is
\(\mathrm{d}^{f}\) Link_output_delay over \(\mathcal{O}\) — snapshots a port's delivery lags its link's flow (PyPSA's delay, delay2, … read long). In data prep you divide it by the snapshot_weightings.generators value and round it up. It is zero for a port that delivers at once. A port has the same delay in every scenario, because PyPSA refuses a delay that differs by scenario (constants.py:55)
\(\overline{\mathrm{f}}\) Link_p_max_pu over \(\mathcal{L} \times \Xi \times \mathcal{T}\) — most flow, per unit of nominal power
\(\underline{\mathrm{f}}\) Link_p_min_pu over \(\mathcal{L} \times \Xi \times \mathcal{T}\) — least flow, per unit of nominal power: negative for a link that carries both ways
\(\mathrm{f}^{\mathrm{nom}}\) Link_p_nom over \(\mathcal{L} \times \Xi\) — nominal power
\(\mathrm{ext}^{f}\) Link_p_nom_extendable over \(\mathcal{L}\) — whether the nominal power is a decision
\(\mathrm{on}^{\mathrm{load}}\) Load_active over \(\mathcal{D}\) — whether a load is active in the network, PyPSA's active. A load has no build year and no lifetime, so the flag is the same in every snapshot. PyPSA refuses one that differs by scenario (constants.py:54)
\(\mathrm{load}\) Load_p_set over \(\mathcal{D} \times \Xi \times \mathcal{T}\) — demand
\(\mathrm{sgn}^{\mathrm{load}}\) Load_sign over \(\mathcal{D}\) — the sign a load's demand enters its bus's balance with. It is PyPSA's sign, -1 unless given, 1 for a load that feeds its bus. PyPSA refuses one that differs by scenario (constants.py:46)
\(\mathrm{on}^{z}\) Process_active over \(\mathcal{J} \times \mathcal{T}\) — whether a process is active in a snapshot's period. It is PyPSA's active, data prep
\(\mathrm{cyc}^{z}\) Process_output_cyclic_delay over \(\mathcal{R}\) — whether a delayed port's transfer wraps from the end of its investment period (PyPSA's cyclic_delay0, cyclic_delay1, …). Where it does not, the energy still in transit at each period's first snapshots is lost. The flag is the same in every scenario, as the delay is
\(\mathrm{d}^{z}\) Process_output_delay over \(\mathcal{R}\) — snapshots a port's transfer lags its process's internal power (PyPSA's delay0, delay1, … read long). In data prep you divide it by the snapshot_weightings.generators value and round it up. It is zero for a port that transfers at once. A port has the same delay in every scenario, as a link's port has
\(\alpha\) Process_rate over \(\mathcal{R} \times \Xi \times \mathcal{T}\) — the energy a port draws or delivers per unit of internal power (PyPSA's rate0, rate1, … read long). It is negative where the port withdraws and positive where it injects. A link is a process whose bus0 rate is minus one and whose output rates are its efficiencies. The rate at the snapshot the transfer arrives applies, so a delayed port transfers at its arrival snapshot's rate (constraints.py:1575)
\(\mathrm{w}^{y}\) period_weight_objective over \(\mathcal{Y}\) — PyPSA's investment_period_weightings.objective — what a period's cost weighs. PyPSA reads it only under multi_investment_periods. Otherwise you feed one in data prep, whatever the column contains (optimize.py:231-233, :293-297)
\(\pi\) scenario_weight over \(\Xi\) — PyPSA's scenario_weightings.weight — the probability of a future
\(\mathrm{w}\) snapshot_weightings_objective over \(\mathcal{T}\) — PyPSA's snapshot_weightings.objective — hours a snapshot weighs in the cost

Conditions

Symbol Meaning
\(\mathrm{Generator\_marginal\_cost\_curve}\) Generator_marginal_cost_curve over \(\mathcal{G}\) — a generator that pays along a piecewise marginal_cost curve: one of PyPSA's active_assets with curve data (optimize.py:307, piecewise.py:230-257)

Variables

Symbol Meaning
\(CVaR\) CVaR (scalar), \(0\) where the mask leaves it out — CVaR — the tail's average cost, what the objective prices at omega
\(a\) CVaR_a over \(\Xi\), \(0\) where the mask leaves it out — CVaR-a — how far a scenario's operating cost exceeds the tail's start. It is nothing where it does not
\(\theta\) CVaR_theta (scalar), \(0\) where the mask leaves it out — CVaR-theta — where the tail starts, the value at risk
\(C^{\mathrm{curve}}\) Generator_marginal_cost_piecewise over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — Generator-marginal_cost_piecewise — the running cost of a generator that pays along a curve, held above the curve
\(p\) Generator_p over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — Generator-p — output of a generator in a snapshot
\(P\) Generator_p_nom_ext over \(\mathcal{G}\) — Generator-p_nom — nominal power where it is a decision, for a generator that is active. A fixed build is the parameter of the same PyPSA name
\(f\) Link_p over \(\mathcal{L} \times \Xi \times \mathcal{T}\) — Link-p — PyPSA's p0, the flow measured at the Link_bus0 end. A positive value withdraws there and injects at every bus the link's output ports deliver to
\(z\) Process_p over \(\mathcal{J} \times \Xi \times \mathcal{T}\) — Process-p — PyPSA's internal power p. A positive value drives every port at its own rate, withdrawing where the rate is negative and injecting where it is positive

Definitions

Symbol Meaning
\(\mathit{Bus\_injection}\) Bus_injection over \(\mathcal{N} \times \Xi \times \mathcal{T}\) — what every component puts into a bus, less what it takes out of it. PyPSA writes each term into the balance, and a load on its right-hand side
\(\mathit{Generator\_capex}\) Generator_capex (scalar)
\(\mathit{Generator\_injection}\) Generator_injection over \(\mathcal{N} \times \Xi \times \mathcal{T}\)
\(\mathit{Generator\_opex}\) Generator_opex over \(\Xi\)
\(\mathit{Generator\_running\_cost}\) Generator_running_cost over \(\mathcal{G} \times \Xi \times \mathcal{T}\) — what a generator's output costs in a snapshot: the value read off its piecewise curve where it has one, and its linear marginal_cost otherwise. A curve replaces the linear cost, as PyPSA drops the unit from that term (optimize.py:333-348)
\(\mathit{Link\_injection}\) Link_injection over \(\mathcal{N} \times \Xi \times \mathcal{T}\)
\(\mathit{Link\_opex}\) Link_opex over \(\Xi\)
\(\mathit{Link\_output\_arrival}\) Link_output_arrival over \(\mathcal{O} \times \Xi \times \mathcal{T}\) — what a link delivers to an output port at a snapshot. It is its flow delayed by the port's delay within its investment period. It is multiplied by the port's efficiency at the snapshot the flow arrives. Where the port is cyclic_delay, the delayed flow wraps from the period's end. Where it is not, the flow still in transit at the period's first snapshots is lost. A port that does not delay (delay zero) delivers its flow unshifted, cyclic or not
\(\mathrm{Load\_demand}\) Load_demand over \(\mathcal{D} \times \Xi \times \mathcal{T}\) — what a load draws from its bus's balance. It is its demand times its sign where it is active, and nothing where it is not. PyPSA drops an inactive load from the balance (constraints.py:1591-1592)
\(\mathrm{Load\_injection}\) Load_injection over \(\mathcal{N} \times \Xi \times \mathcal{T}\)
\(\mathit{Process\_injection}\) Process_injection over \(\mathcal{N} \times \Xi \times \mathcal{T}\)
\(\mathit{Process\_output\_arrival}\) Process_output_arrival over \(\mathcal{R} \times \Xi \times \mathcal{T}\) — what a process transfers at a port at a snapshot. It is its internal power delayed by the port's delay within its investment period. It is multiplied by the port's rate at the snapshot the transfer arrives. Where the port is cyclic_delay, the delayed transfer wraps from the period's end. Where it is not, the energy still in transit at the period's first snapshots is lost. A port that does not delay (delay zero) transfers at once, cyclic or not
\(\mathit{risk\_weighted\_opex}\) risk_weighted_opex (scalar)
\(\mathit{scenario\_opex}\) scenario_opex over \(\Xi\) — what a future costs to run: every operating term, weighted by the snapshot's hours and its period, before the scenario's own weight. A start and a stop cost what they cost, unweighted, as PyPSA adds them (optimize.py:448-468)
\(\mathit{total\_cost}\) total_cost (scalar) — what the system costs. Capacity counts once per active period at its expected cost over the scenarios. Operation counts in expectation over the scenarios, and a share of it at the tail

Notation

Symbol Meaning
\(t \ominus k\) edge='wrap' — \(t-k\), cyclic
\(t \boxminus_{v} k\) edge=v — \(t-k\), and \(v\) past the edge
\(t \ominus^{\mathrm{relation}(t)} k\) within=relation[...] — \(t-k\) inside the group of \(t\)

Conditions

Generator_marginal_cost_curve

\[ \mathrm{Generator\_marginal\_cost\_curve}_{g} \iff \lvert \{ b \in \mathcal{B} \,:\, \mathrm{p}^{\mathrm{curve}}_{b,g} \text{ is defined} \} \rvert > 0 \wedge \mathrm{act}_{g} \qquad \forall\, g \in \mathcal{G} \]

Variable domains

CVaR

\[ CVaR \in \mathbb{R} \qquad \text{where } \omega \text{ is defined} \]

CVaR_a

\[ a_{\xi} \ge 0 \qquad \forall\, \xi \in \Xi \,:\, \omega \text{ is defined} \]

CVaR_theta

\[ \theta \in \mathbb{R} \qquad \text{where } \omega \text{ is defined} \]

Generator_marginal_cost_piecewise

\[ C^{\mathrm{curve}}_{g,\xi,t} \in \mathbb{R} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{Generator\_marginal\_cost\_curve}_{g} \wedge \mathrm{on}_{g,t} \]

Generator_p

\[ p_{g,\xi,t} \in \mathbb{R} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{on}_{g,t} \]

Generator_p_nom_ext

\[ P_{g} \in \mathbb{R} \qquad \forall\, g \in \mathcal{G} \,:\, \mathrm{ext}_{g} \wedge \mathrm{act}_{g} \]

Link_p

\[ f_{l,\xi,t} \in \mathbb{R} \qquad \forall\, l \in \mathcal{L},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{on}^{f}_{l,t} \]

Process_p

\[ z_{j,\xi,t} \in \mathbb{R} \qquad \forall\, j \in \mathcal{J},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{on}^{z}_{j,t} \]

Definitions

Bus_injection

\[ \mathit{Bus\_injection}_{n,\xi,t} = \mathit{Generator\_injection}_{n,\xi,t} + \mathit{Link\_injection}_{n,\xi,t} + \mathrm{Load\_injection}_{n,\xi,t} + \mathit{Process\_injection}_{n,\xi,t} \qquad \forall\, n \in \mathcal{N},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Generator_capex

\[ \mathit{Generator\_capex} = \sum_{g \in \mathcal{G},\ \xi \in \Xi} \pi_{\xi} \cdot P_{g} \cdot \mathrm{c}^{\mathrm{cap}}_{g,\xi} \cdot \mathrm{W}_{g} \]

Generator_injection

\[ \mathit{Generator\_injection}_{n,\xi,t} = \sum_{g \in \mathcal{G} \,:\, \mathrm{Generator\_bus}(g) = n} \mathrm{sgn}_{g} \cdot p_{g,\xi,t} \qquad \forall\, n \in \mathcal{N},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Generator_opex

\[ \mathit{Generator\_opex}_{\xi} = \sum_{t \in \mathcal{T}} \sum_{g \in \mathcal{G}} \mathit{Generator\_running\_cost}_{g,\xi,t} \cdot \mathrm{w}_{t} \cdot \mathrm{w}^{y}_{\mathrm{snapshot\_period}(t)} + \sum_{t \in \mathcal{T}} \sum_{g \in \mathcal{G}} p_{g,\xi,t} \cdot p_{g,\xi,t} \cdot \mathrm{c}^{(2)}_{g,\xi,t} \cdot \mathrm{w}_{t} \cdot \mathrm{w}^{y}_{\mathrm{snapshot\_period}(t)} \qquad \forall\, \xi \in \Xi \]

Generator_running_cost

\[ \mathit{Generator\_running\_cost}_{g,\xi,t} = \begin{cases} C^{\mathrm{curve}}_{g,\xi,t} & \text{if } \mathrm{Generator\_marginal\_cost\_curve}_{g} \\ p_{g,\xi,t} \cdot \mathrm{c}_{g,\xi,t} & \text{otherwise} \end{cases} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Link_injection

\[ \mathit{Link\_injection}_{n,\xi,t} = -\left( \sum_{l \in \mathcal{L} \,:\, \mathrm{Link\_bus0}(l) = n} f_{l,\xi,t} \right) + \sum_{o \in \mathcal{O} \,:\, \mathrm{Link\_output\_bus}(o) = n} \mathit{Link\_output\_arrival}_{o,\xi,t} \qquad \forall\, n \in \mathcal{N},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Link_opex

\[ \mathit{Link\_opex}_{\xi} = \sum_{t \in \mathcal{T}} \sum_{l \in \mathcal{L}} f_{l,\xi,t} \cdot \mathrm{c}^{f}_{l,\xi,t} \cdot \mathrm{w}_{t} \cdot \mathrm{w}^{y}_{\mathrm{snapshot\_period}(t)} + \sum_{t \in \mathcal{T}} \sum_{l \in \mathcal{L}} f_{l,\xi,t} \cdot f_{l,\xi,t} \cdot \mathrm{c}^{f,(2)}_{l,\xi,t} \cdot \mathrm{w}_{t} \cdot \mathrm{w}^{y}_{\mathrm{snapshot\_period}(t)} \qquad \forall\, \xi \in \Xi \]

Link_output_arrival

\[ \mathit{Link\_output\_arrival}_{o,\xi,t} = \begin{cases} f_{\mathrm{Link\_output\_link}(o),\xi,t \ominus^{\mathrm{snapshot\_period}(t)} \mathrm{d}^{f}} \cdot \eta_{o,\xi,t} & \text{if } \mathrm{cyc}^{f}_{o} \\ f_{\mathrm{Link\_output\_link}(o),\xi,t \boxminus_{0}^{\mathrm{snapshot\_period}(t)} \mathrm{d}^{f}} \cdot \eta_{o,\xi,t} & \text{otherwise} \end{cases} \qquad \forall\, o \in \mathcal{O},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Load_demand

\[ \mathrm{Load\_demand}_{d,\xi,t} = \begin{cases} \mathrm{sgn}^{\mathrm{load}}_{d} \cdot \mathrm{load}_{d,\xi,t} & \text{if } \mathrm{on}^{\mathrm{load}}_{d} \\ 0 & \text{otherwise} \end{cases} \qquad \forall\, d \in \mathcal{D},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Load_injection

\[ \mathrm{Load\_injection}_{n,\xi,t} = \sum_{d \in \mathcal{D} \,:\, \mathrm{Load\_bus}(d) = n} \mathrm{Load\_demand}_{d,\xi,t} \qquad \forall\, n \in \mathcal{N},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Process_injection

\[ \mathit{Process\_injection}_{n,\xi,t} = \sum_{r \in \mathcal{R} \,:\, \mathrm{Process\_output\_bus}(r) = n} \mathit{Process\_output\_arrival}_{r,\xi,t} \qquad \forall\, n \in \mathcal{N},\ \xi \in \Xi,\ t \in \mathcal{T} \]

Process_output_arrival

\[ \mathit{Process\_output\_arrival}_{r,\xi,t} = \begin{cases} z_{\mathrm{Process\_output\_process}(r),\xi,t \ominus^{\mathrm{snapshot\_period}(t)} \mathrm{d}^{z}} \cdot \alpha_{r,\xi,t} & \text{if } \mathrm{cyc}^{z}_{r} \\ z_{\mathrm{Process\_output\_process}(r),\xi,t \boxminus_{0}^{\mathrm{snapshot\_period}(t)} \mathrm{d}^{z}} \cdot \alpha_{r,\xi,t} & \text{otherwise} \end{cases} \qquad \forall\, r \in \mathcal{R},\ \xi \in \Xi,\ t \in \mathcal{T} \]

risk_weighted_opex

\[ \mathit{risk\_weighted\_opex} = \left( 1 - \omega \right) \cdot \left( \sum_{\xi \in \Xi} \pi_{\xi} \cdot \mathit{scenario\_opex}_{\xi} \right) + \omega \cdot CVaR \]

scenario_opex

\[ \mathit{scenario\_opex}_{\xi} = \mathit{Generator\_opex}_{\xi} + \mathit{Link\_opex}_{\xi} \qquad \forall\, \xi \in \Xi \]

total_cost

\[ \mathit{total\_cost} = \mathit{Generator\_capex} + \mathit{risk\_weighted\_opex} \]

Objective

\[ \min \mathit{total\_cost} \]

Subject to

Bus_nodal_balance

\[ \mathit{Bus\_injection}_{n,\xi,t} = 0 \qquad \forall\, n \in \mathcal{N},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{Bus\_attached}_{n,t} \]

CVaR_def

\[ \theta + \frac{1}{1 - \alpha} \cdot \left( \sum_{\xi \in \Xi} \pi_{\xi} \cdot a_{\xi} \right) \le CVaR \qquad \text{where } \omega \text{ is defined} \]

CVaR_excess

\[ a_{\xi} - \mathit{scenario\_opex}_{\xi} + \theta \ge 0 \qquad \forall\, \xi \in \Xi \,:\, \omega \text{ is defined} \]

Generator_ext_p_lower

\[ p_{g,\xi,t} \ge \underline{\mathrm{p}}_{g,\xi,t} \cdot P_{g} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{ext}_{g} \wedge \neg \mathrm{com}_{g} \wedge \mathrm{on}_{g,t} \]

Generator_ext_p_nom_lower

\[ P_{g} \ge \underline{\mathrm{p}}^{\mathrm{nom}}_{g,\xi} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi \,:\, \mathrm{ext}_{g} \wedge \mathrm{act}_{g} \]

Generator_ext_p_nom_upper

\[ P_{g} \le \overline{\mathrm{p}}^{\mathrm{nom}}_{g,\xi} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi \,:\, \mathrm{ext}_{g} \wedge \mathrm{act}_{g} \wedge \overline{\mathrm{p}}^{\mathrm{nom}}_{g,\xi} \text{ is defined} \]

Generator_ext_p_upper

\[ p_{g,\xi,t} \le \overline{\mathrm{p}}_{g,\xi,t} \cdot P_{g} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \mathrm{ext}_{g} \wedge \neg \mathrm{com}_{g} \wedge \mathrm{on}_{g,t} \]

Generator_fix_p_lower

\[ p_{g,\xi,t} \ge \underline{\mathrm{p}}_{g,\xi,t} \cdot \mathrm{p}^{\mathrm{nom}}_{g,\xi} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \neg \mathrm{ext}_{g} \wedge \neg \mathrm{com}_{g} \wedge \mathrm{on}_{g,t} \]

Generator_fix_p_upper

\[ p_{g,\xi,t} \le \overline{\mathrm{p}}_{g,\xi,t} \cdot \mathrm{p}^{\mathrm{nom}}_{g,\xi} \qquad \forall\, g \in \mathcal{G},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \neg \mathrm{ext}_{g} \wedge \neg \mathrm{com}_{g} \wedge \mathrm{on}_{g,t} \]

Link_fix_p_lower

\[ f_{l,\xi,t} \ge \underline{\mathrm{f}}_{l,\xi,t} \cdot \mathrm{f}^{\mathrm{nom}}_{l,\xi} \qquad \forall\, l \in \mathcal{L},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \neg \mathrm{ext}^{f}_{l} \wedge \neg \mathrm{com}^{f}_{l} \wedge \mathrm{on}^{f}_{l,t} \]

Link_fix_p_upper

\[ f_{l,\xi,t} \le \overline{\mathrm{f}}_{l,\xi,t} \cdot \mathrm{f}^{\mathrm{nom}}_{l,\xi} \qquad \forall\, l \in \mathcal{L},\ \xi \in \Xi,\ t \in \mathcal{T} \,:\, \neg \mathrm{ext}^{f}_{l} \wedge \neg \mathrm{com}^{f}_{l} \wedge \mathrm{on}^{f}_{l,t} \]

The spec, differential/pypsa/rungs/rung_14_stochastic.yaml — the file projected onto what this rung builds:

version: 0
description: The file states a plain `n.optimize()`, its multi-period class under the `multi_investment_periods`
  keyword, and its stochastic class, which `n.set_scenarios(...)` sets. Every quantity chosen once the
  future is known (second-stage) spans the futures dispatch is chosen in (`scenario`). Capacity is chosen
  once, before the future is known, and paid at its expected cost in each investment `period` its build
  year and lifetime span. Operation is the expectation over the scenarios' weights, with a share priced
  at the worst-outcome mean cost (CVaR) only where a risk preference is set. A plain run's data has one
  scenario, one period, unit weights, no outage and every `active` flag true, so the problem is the standard
  one. On a security-constrained run, PyPSA copies each branch flow limit once per `outage`, and the outage
  factors, `active` flags and scenario weights come from data prep.
dimensions:
  breakpoint: {dtype: int, ordered: true, description: 'the corners of a piecewise-linear curve, in order.
      A curve runs over the breakpoints its data has rows for, so curves may differ in length'}
  bus: {description: network nodes}
  generator: {description: 'generating units, each on one bus'}
  link: {description: 'controllable connections, each from one bus to the buses it delivers to'}
  link_output: {description: 'a link''s output ports, one label per port a link declares. The labels are
      PyPSA''s `bus1`, `bus2`, … columns read long, data prep. A link of any number of output ports is
      then one term in the balance'}
  load: {description: 'demands, each on one bus'}
  period: {dtype: int, ordered: true, description: investment periods. It is PyPSA's `investment_periods`}
  process: {description: 'generalised multi-port converters, each with an internal power that every port
      draws or delivers at its own rate'}
  process_output: {description: 'a process''s ports, one label per port a process declares. The labels
      are PyPSA''s `bus0`, `bus1`, …, each with a signed `rate`, data prep. A process of any number of
      ports is then one term in the balance'}
  scenario: {description: 'the futures dispatch is chosen in, each with a weight'}
  snapshot: {dtype: datetime, ordered: true, description: dispatch periods}
relations:
  Generator_bus: {key: generator, values: bus, description: the bus a generator sits on}
  Link_bus0: {key: link, values: bus, description: the bus a link leaves}
  Link_output_bus: {key: link_output, values: bus, description: 'the bus an output port delivers to. It
      is PyPSA''s `bus1`, `bus2`, … columns. A link of three output ports is three labels here rather
      than a third relation, so the file states any number of them'}
  Link_output_link: {key: link_output, values: link, description: the link an output port belongs to}
  Load_bus: {key: load, values: bus, description: the bus a load sits on}
  Process_output_bus: {key: process_output, values: bus, description: 'the bus a port draws from or delivers
      to. It is PyPSA''s `bus0`, `bus1`, … columns. A process of three ports is three labels here rather
      than a third relation, so the file states any number of them'}
  Process_output_process: {key: process_output, values: process, description: the process a port belongs
      to}
  snapshot_period: {key: snapshot, values: period, description: the investment period a snapshot falls
      in}
parameters:
  Bus_attached:
    dims: [bus, snapshot]
    dtype: bool
    description: whether a load, a unit, a port or a branch that is active in the snapshot's period is
      attached to the bus, data prep. A bus nothing is attached to has no balance row
  CVaR_alpha:
    dims: []
    missing: absent
    description: PyPSA's `risk_preference['alpha']` — the confidence level. The tail is the other `1 -
      alpha` of the probability. `CVaR_alpha` has a value exactly where `CVaR_omega` has one
  CVaR_omega:
    dims: []
    missing: 0.0
    description: PyPSA's `risk_preference['omega']` — the share of operating cost priced at the tail rather
      than in expectation. `CVaR_omega` has a value only where a risk preference is set. Without a risk
      preference, the file has no CVaR variables or rows. With `omega = 0`, PyPSA builds them, and the
      optimum is the risk-neutral one
  Generator_active:
    dims: [generator, snapshot]
    dtype: bool
    description: whether a generator is active in a snapshot's period. It is PyPSA's `active`, from build
      year and lifetime, data prep
  Generator_active_asset:
    dims: [generator]
    dtype: bool
    description: whether a generator is one of PyPSA's `active_assets`, by its `active` flag alone, without
      build year and lifetime. A unit that is active in no snapshot of the horizon still counts. PyPSA
      refuses one that differs by scenario (`constants.py:54`)
  Generator_capital_cost:
    dims: [generator, scenario]
    description: cost of one unit of nominal power for the modelled horizon (PyPSA's `periodized_cost`),
      data prep. It is `overnight_cost` as an annuity over `lifetime` at `discount_rate`, times `nyears`,
      where `overnight_cost` is given, and `capital_cost` where it is not. Both add `fom_cost` (`components.py:1157-1178`,
      `costs.py:102-203`)
  Generator_capital_weight:
    dims: [generator]
    description: the sum of period weights a generator is active in. It is PyPSA's `active * period_weighting`,
      summed, data prep
  Generator_committable:
    dims: [generator]
    dtype: bool
    description: whether output is gated by an on/off status decision
  Generator_marginal_cost:
    dims: [generator, scenario, snapshot]
    description: cost of one unit of output
  Generator_marginal_cost_curve_p:
    dims: [breakpoint, generator]
    missing: absent
    description: 'the output at each breakpoint of a generator''s piecewise `marginal_cost` curve: PyPSA''s
      `p_pu` breakpoints times `p_nom`, data prep. A generator with no curve has no rows'
  Generator_marginal_cost_quadratic:
    dims: [generator, scenario, snapshot]
    description: cost of the square of one unit of output
  Generator_p_max_pu:
    dims: [generator, scenario, snapshot]
    description: 'most output, per unit of nominal power: an availability profile'
  Generator_p_min_pu:
    dims: [generator, scenario, snapshot]
    description: least output, per unit of nominal power
  Generator_p_nom:
    dims: [generator, scenario]
    description: nominal power
  Generator_p_nom_extendable:
    dims: [generator]
    dtype: bool
    description: whether the nominal power is a decision
  Generator_p_nom_max:
    dims: [generator, scenario]
    description: most nominal power an extendable generator may be built at
  Generator_p_nom_min:
    dims: [generator, scenario]
    description: least nominal power an extendable generator may be built at
  Generator_sign:
    dims: [generator]
    description: the sign output enters its bus's balance with. It is PyPSA's `sign`, `1` unless given,
      `-1` for a unit that draws power. PyPSA refuses one that differs by scenario (`constants.py:46`)
  Link_active:
    dims: [link, snapshot]
    dtype: bool
    description: whether a link is active in a snapshot's period. It is PyPSA's `active`, data prep
  Link_committable:
    dims: [link]
    dtype: bool
    description: whether flow is gated by an on/off status decision
  Link_efficiency:
    dims: [link_output, scenario, snapshot]
    description: 'share of the flow that arrives at an output port, PyPSA''s `efficiency`, `efficiency2`,
      … read long: negative where that port consumes rather than delivers. Read at the snapshot the flow
      arrives, so a delayed port delivers at its arrival snapshot''s efficiency (`constraints.py:1575`)'
  Link_marginal_cost:
    dims: [link, scenario, snapshot]
    description: cost of one unit of flow
  Link_marginal_cost_quadratic:
    dims: [link, scenario, snapshot]
    description: cost of the square of one unit of flow
  Link_output_cyclic_delay:
    dims: [link_output]
    dtype: bool
    description: whether a delayed port's flow wraps from the end of its investment period (PyPSA's `cyclic_delay`,
      `cyclic_delay2`, …). Where it does not, the flow still in transit at each period's first snapshots
      is lost. The flag is the same in every scenario, as the delay is
  Link_output_delay:
    dims: [link_output]
    dtype: int
    description: snapshots a port's delivery lags its link's flow (PyPSA's `delay`, `delay2`, … read long).
      In data prep you divide it by the `snapshot_weightings.generators` value and round it up. It is
      zero for a port that delivers at once. A port has the same delay in every scenario, because PyPSA
      refuses a delay that differs by scenario (`constants.py:55`)
  Link_p_max_pu:
    dims: [link, scenario, snapshot]
    description: most flow, per unit of nominal power
  Link_p_min_pu:
    dims: [link, scenario, snapshot]
    description: 'least flow, per unit of nominal power: negative for a link that carries both ways'
  Link_p_nom:
    dims: [link, scenario]
    description: nominal power
  Link_p_nom_extendable:
    dims: [link]
    dtype: bool
    description: whether the nominal power is a decision
  Load_active:
    dims: [load]
    dtype: bool
    description: whether a load is active in the network, PyPSA's `active`. A load has no build year and
      no lifetime, so the flag is the same in every snapshot. PyPSA refuses one that differs by scenario
      (`constants.py:54`)
  Load_p_set:
    dims: [load, scenario, snapshot]
    description: demand
  Load_sign:
    dims: [load]
    description: the sign a load's demand enters its bus's balance with. It is PyPSA's `sign`, `-1` unless
      given, `1` for a load that feeds its bus. PyPSA refuses one that differs by scenario (`constants.py:46`)
  Process_active:
    dims: [process, snapshot]
    dtype: bool
    description: whether a process is active in a snapshot's period. It is PyPSA's `active`, data prep
  Process_output_cyclic_delay:
    dims: [process_output]
    dtype: bool
    description: whether a delayed port's transfer wraps from the end of its investment period (PyPSA's
      `cyclic_delay0`, `cyclic_delay1`, …). Where it does not, the energy still in transit at each period's
      first snapshots is lost. The flag is the same in every scenario, as the delay is
  Process_output_delay:
    dims: [process_output]
    dtype: int
    description: snapshots a port's transfer lags its process's internal power (PyPSA's `delay0`, `delay1`,
      … read long). In data prep you divide it by the `snapshot_weightings.generators` value and round
      it up. It is zero for a port that transfers at once. A port has the same delay in every scenario,
      as a link's port has
  Process_rate:
    dims: [process_output, scenario, snapshot]
    description: the energy a port draws or delivers per unit of internal power (PyPSA's `rate0`, `rate1`,
      … read long). It is negative where the port withdraws and positive where it injects. A link is a
      process whose `bus0` rate is minus one and whose output rates are its efficiencies. The rate at
      the snapshot the transfer arrives applies, so a delayed port transfers at its arrival snapshot's
      rate (`constraints.py:1575`)
  period_weight_objective:
    dims: [period]
    description: PyPSA's `investment_period_weightings.objective` — what a period's cost weighs. PyPSA
      reads it only under `multi_investment_periods`. Otherwise you feed one in data prep, whatever the
      column contains (`optimize.py:231-233`, `:293-297`)
  scenario_weight:
    dims: [scenario]
    description: PyPSA's `scenario_weightings.weight` — the probability of a future
  snapshot_weightings_objective:
    dims: [snapshot]
    description: PyPSA's `snapshot_weightings.objective` — hours a snapshot weighs in the cost
variables:
  CVaR:
    dims: []
    where: CVaR_omega
    missing: 0.0
    description: '`CVaR` — the tail''s average cost, what the objective prices at `omega`'
  CVaR_a:
    dims: [scenario]
    where: CVaR_omega
    bounds: {lower: 0.0}
    missing: 0.0
    description: '`CVaR-a` — how far a scenario''s operating cost exceeds the tail''s start. It is nothing
      where it does not'
  CVaR_theta:
    dims: []
    where: CVaR_omega
    missing: 0.0
    description: '`CVaR-theta` — where the tail starts, the value at risk'
  Generator_p:
    dims: [generator, scenario, snapshot]
    where: Generator_active
    description: '`Generator-p` — output of a generator in a snapshot'
  Generator_p_nom_ext:
    dims: [generator]
    where: Generator_p_nom_extendable AND Generator_active_asset
    description: '`Generator-p_nom` — nominal power where it is a decision, for a generator that is active.
      A fixed build is the parameter of the same PyPSA name'
  Link_p:
    dims: [link, scenario, snapshot]
    where: Link_active
    description: '`Link-p` — PyPSA''s `p0`, the flow measured at the `Link_bus0` end. A positive value
      withdraws there and injects at every bus the link''s output ports deliver to'
  Generator_marginal_cost_piecewise:
    dims: [generator, scenario, snapshot]
    where: Generator_marginal_cost_curve AND Generator_active
    description: '`Generator-marginal_cost_piecewise` — the running cost of a generator that pays along
      a curve, held above the curve'
  Process_p:
    dims: [process, scenario, snapshot]
    where: Process_active
    description: '`Process-p` — PyPSA''s internal power `p`. A positive value drives every port at its
      own rate, withdrawing where the rate is negative and injecting where it is positive'
constraints:
  Bus_nodal_balance:
    dims: [bus, scenario, snapshot]
    where: Bus_attached
    expression: Bus_injection == 0
    description: '`Bus-nodal_balance` — what is generated at a bus meets the load there, less half of
      every incident line''s and transformer''s loss. What is generated includes storage dispatch and
      stores. It is net of what the links take away, plus what arrives over them after losses and any
      delay at every port they deliver to. Each process port draws or delivers at its own rate, and each
      passive branch carries its flow. PyPSA dissipates a branch''s loss half at either end. Each generator,
      storage unit, store and load term enters with its component''s `sign` (`constraints.py:1480-1481`,
      `:1592`), and an inactive load not at all. A bus nothing is attached to has no row. PyPSA refuses
      one that carries load. The file states no assumption for it.'
  CVaR_def:
    dims: []
    where: CVaR_omega
    expression: CVaR_theta + 1 / (1 - CVaR_alpha) * sum(scenario_weight * CVaR_a, over=scenario) <= CVaR
    description: '`CVaR-def` — the tail''s average is at least where it starts plus the expected excess
      over the tail''s probability'
  CVaR_excess:
    dims: [scenario]
    where: CVaR_omega
    expression: CVaR_a - scenario_opex + CVaR_theta >= 0
    description: '`CVaR-excess-{s}` — a scenario''s operating cost beyond the tail''s start is its excess.
      PyPSA names one row per scenario'
  Generator_ext_p_lower:
    dims: [generator, scenario, snapshot]
    where: Generator_p_nom_extendable AND not Generator_committable AND Generator_active
    expression: Generator_p >= Generator_p_min_pu * Generator_p_nom_ext
    description: '`Generator-ext-p-lower` — an extendable generator outputs at least its minimum of the
      chosen build'
  Generator_ext_p_nom_lower:
    dims: [generator, scenario]
    where: Generator_p_nom_extendable AND Generator_active_asset
    expression: Generator_p_nom_ext >= Generator_p_nom_min
    description: '`Generator-ext-p_nom-lower` — the chosen build is at least its floor in every scenario'
  Generator_ext_p_nom_upper:
    dims: [generator, scenario]
    where: Generator_p_nom_extendable AND Generator_active_asset AND Generator_p_nom_max
    expression: Generator_p_nom_ext <= Generator_p_nom_max
    description: '`Generator-ext-p_nom-upper` — the chosen build is at most its cap in every scenario.
      A cap of infinity is no row'
  Generator_ext_p_upper:
    dims: [generator, scenario, snapshot]
    where: Generator_p_nom_extendable AND not Generator_committable AND Generator_active
    expression: Generator_p <= Generator_p_max_pu * Generator_p_nom_ext
    description: '`Generator-ext-p-upper` — an extendable generator outputs at most what is available
      of the chosen build'
  Generator_fix_p_lower:
    dims: [generator, scenario, snapshot]
    where: not Generator_p_nom_extendable AND not Generator_committable AND Generator_active
    expression: Generator_p >= Generator_p_min_pu * Generator_p_nom
    description: '`Generator-fix-p-lower` — a fixed generator outputs at least its minimum'
  Generator_fix_p_upper:
    dims: [generator, scenario, snapshot]
    where: not Generator_p_nom_extendable AND not Generator_committable AND Generator_active
    expression: Generator_p <= Generator_p_max_pu * Generator_p_nom
    description: '`Generator-fix-p-upper` — a fixed generator outputs at most what is available'
  Link_fix_p_lower:
    dims: [link, scenario, snapshot]
    where: not Link_p_nom_extendable AND not Link_committable AND Link_active
    expression: Link_p >= Link_p_min_pu * Link_p_nom
    description: '`Link-fix-p-lower` — a fixed link carries at least its minimum, negative for the other
      way'
  Link_fix_p_upper:
    dims: [link, scenario, snapshot]
    where: not Link_p_nom_extendable AND not Link_committable AND Link_active
    expression: Link_p <= Link_p_max_pu * Link_p_nom
    description: '`Link-fix-p-upper` — a fixed link carries at most its nominal power'
expressions:
  Bus_injection:
    dims: [bus, scenario, snapshot]
    expression: ((Generator_injection + Link_injection) + Load_injection) + Process_injection
    description: what every component puts into a bus, less what it takes out of it. PyPSA writes each
      term into the balance, and a load on its right-hand side
  Generator_capex: {expression: sum(scenario_weight * Generator_p_nom_ext * Generator_capital_cost * Generator_capital_weight)}
  Generator_injection: {expression: 'sum(Generator_sign * Generator_p, over=generator, by=Generator_bus[bus])'}
  Generator_opex: {expression: 'sum(sum((Generator_running_cost * snapshot_weightings_objective) * at(period_weight_objective,
      by=snapshot_period[period]), over=generator), over=snapshot) + sum(sum((((Generator_p * Generator_p)
      * Generator_marginal_cost_quadratic) * snapshot_weightings_objective) * at(period_weight_objective,
      by=snapshot_period[period]), over=generator), over=snapshot)'}
  Generator_running_cost:
    dims: [generator, scenario, snapshot]
    description: 'what a generator''s output costs in a snapshot: the value read off its piecewise curve
      where it has one, and its linear `marginal_cost` otherwise. A curve replaces the linear cost, as
      PyPSA drops the unit from that term (`optimize.py:333-348`)'
    cases:
      curve: {when: Generator_marginal_cost_curve, expression: Generator_marginal_cost_piecewise}
    otherwise: Generator_p * Generator_marginal_cost
  Link_injection: {expression: '-sum(Link_p, over=link, by=Link_bus0[bus]) + sum(Link_output_arrival,
      over=link_output, by=Link_output_bus[bus])'}
  Link_opex: {expression: 'sum(sum(((Link_p * Link_marginal_cost) * snapshot_weightings_objective) * at(period_weight_objective,
      by=snapshot_period[period]), over=link), over=snapshot) + sum(sum((((Link_p * Link_p) * Link_marginal_cost_quadratic)
      * snapshot_weightings_objective) * at(period_weight_objective, by=snapshot_period[period]), over=link),
      over=snapshot)'}
  Link_output_arrival:
    dims: [link_output, scenario, snapshot]
    description: what a link delivers to an output port at a snapshot. It is its flow delayed by the port's
      `delay` within its investment period. It is multiplied by the port's efficiency at the snapshot
      the flow arrives. Where the port is `cyclic_delay`, the delayed flow wraps from the period's end.
      Where it is not, the flow still in transit at the period's first snapshots is lost. A port that
      does not delay (`delay` zero) delivers its flow unshifted, cyclic or not
    cases:
      wrapping: {when: Link_output_cyclic_delay, expression: 'shift(at(Link_p, by=Link_output_link[link]),
          along=snapshot, offset=Link_output_delay, edge=''wrap'', within=snapshot_period[period]) * Link_efficiency'}
    otherwise: shift(at(Link_p, by=Link_output_link[link]), along=snapshot, offset=Link_output_delay,
      edge=0, within=snapshot_period[period]) * Link_efficiency
  Load_demand:
    dims: [load, scenario, snapshot]
    description: what a load draws from its bus's balance. It is its demand times its sign where it is
      active, and nothing where it is not. PyPSA drops an inactive load from the balance (`constraints.py:1591-1592`)
    cases:
      active: {when: Load_active, expression: Load_sign * Load_p_set}
    otherwise: '0'
  Load_injection: {expression: 'sum(Load_demand, over=load, by=Load_bus[bus])'}
  Process_injection: {expression: 'sum(Process_output_arrival, over=process_output, by=Process_output_bus[bus])'}
  Process_output_arrival:
    dims: [process_output, scenario, snapshot]
    description: what a process transfers at a port at a snapshot. It is its internal power delayed by
      the port's `delay` within its investment period. It is multiplied by the port's rate at the snapshot
      the transfer arrives. Where the port is `cyclic_delay`, the delayed transfer wraps from the period's
      end. Where it is not, the energy still in transit at the period's first snapshots is lost. A port
      that does not delay (`delay` zero) transfers at once, cyclic or not
    cases:
      wrapping: {when: Process_output_cyclic_delay, expression: 'shift(at(Process_p, by=Process_output_process[process]),
          along=snapshot, offset=Process_output_delay, edge=''wrap'', within=snapshot_period[period])
          * Process_rate'}
    otherwise: shift(at(Process_p, by=Process_output_process[process]), along=snapshot, offset=Process_output_delay,
      edge=0, within=snapshot_period[period]) * Process_rate
  risk_weighted_opex: {expression: '(1 - CVaR_omega) * sum(scenario_weight * scenario_opex, over=scenario)
      + CVaR_omega * CVaR'}
  scenario_opex:
    dims: [scenario]
    expression: Generator_opex + Link_opex
    description: 'what a future costs to run: every operating term, weighted by the snapshot''s hours
      and its period, before the scenario''s own weight. A start and a stop cost what they cost, unweighted,
      as PyPSA adds them (`optimize.py:448-468`)'
  total_cost:
    dims: []
    expression: Generator_capex + risk_weighted_opex
    description: what the system costs. Capacity counts once per active period at its expected cost over
      the scenarios. Operation counts in expectation over the scenarios, and a share of it at the tail
conditions:
  Generator_marginal_cost_curve: {condition: 'count(Generator_marginal_cost_curve_p, over=breakpoint)
      > 0 AND Generator_active_asset', description: 'a generator that pays along a piecewise `marginal_cost`
      curve: one of PyPSA''s `active_assets` with curve data (`optimize.py:307`, `piecewise.py:230-257`)'}
objective: {sense: minimize, expression: total_cost}

The prep — every table the spec declares, from the network — and the solve:

from differential.pypsa.prep import relation, static, varying, weighting


n = build()  # the network from the PyPSA tab

sources = {
    'bus': pl.Series('bus', list(names(n.buses.index).astype(str)), dtype=pl.String),
        **{
            dim: pl.Series(dim, list(names(n.static(component).index).astype(str)), dtype=pl.String)
            for component, dim in DIM.items()
        },
        **scenarios(n),
        **periods(n, multi),
        **carriers(n, multi),
    'snapshot': pl.Series('snapshot', list(timesteps(n)), dtype=pl.Datetime('us')),
    'Generator_bus': relation(n, 'Generator', 'bus'),
    'Link_bus0': relation(n, 'Link', 'bus0'),
    'Load_bus': relation(n, 'Load', 'bus'),
    'Bus_attached': attached(n),
    'Generator_sign': per_component('Generator', first_scenario(n.generators['sign'])),
    'Load_active': per_component('Load', first_scenario(loads['active']), bool),
    'Load_p_set': varying(n, 'Load', 'p_set'),
    'Load_sign': per_component('Load', first_scenario(loads['sign'])),
    'snapshot_weightings_objective': weighting(n, 'objective'),
}

with sps.solve('differential/pypsa/rungs/rung_14_stochastic.yaml', sources) as solution:
    solution.objective  # 9267.386667

The network, rung_14_stochastic.py in the corpus — the spine plus what this rung adds:

# SPDX-FileCopyrightText: mathspec Contributors
#
# SPDX-License-Identifier: MIT

"""Rung 14: two futures and a risk preference — capacity chosen once, dispatch per scenario."""

from __future__ import annotations

import spine


def build():
    """The spine plus an extendable wind unit whose availability and the south's load differ between a calm and a stormy future."""
    n = spine.build()
    n.add('Generator', 'wind14', bus='south', p_nom_extendable=True, p_nom_max=100, marginal_cost=1, capital_cost=20)
    n.add('Load', 'port14', bus='south')
    n.set_scenarios({'calm': 0.6, 'stormy': 0.4})
    n.c.loads.dynamic.p_set[('calm', 'port14')] = [10, 20, 15, 10]
    n.c.loads.dynamic.p_set[('stormy', 'port14')] = [40, 60, 50, 30]
    n.c.generators.dynamic.p_max_pu[('calm', 'wind14')] = [0.9, 0.7, 0.8, 0.6]
    n.c.generators.dynamic.p_max_pu[('stormy', 'wind14')] = [0.3, 0.2, 0.4, 0.1]
    n.set_risk_preference(alpha=0.5, omega=0.3)
    return n
n = build()
n.optimize(solver_name='highs')
n.objective  # 9267.386667

The data

The tables this rung is the first to declare (2), as the prep produced them:

CVaR_alpha.csv

value
0.5

CVaR_omega.csv

value
0.3