Glossary

Terminology used across this site, with Portuguese equivalents where they appear in Brazilian regulatory or operational documentation. Citations to source papers for the more technical entries live in Bibliography.

Index

Each term links to the section that defines it.

0–9 — 0-order inflow

A — Anticipated dispatch · AR lag

B — Backward pass · Basis · Bellman recursion · Block · Bound stalling · Bus

C — Cascade · CCEE · Census simulation · CEPEL · CLP · Coherent risk measure · Cost scale · Cost-to-go function · Cut (Benders cut) · Cut intercept · Cut pool · Cut slope (cut coefficient) · CVaR · Cycle convergence inequality · Cyclic mode

D — Dead volume · DECOMP · Deficit · DESSEM · Directed acyclic graph (DAG) · Discount factor · Diversion · Downstream · Dual variable · Dynamic cut selection (DCS)

E — EAVaR · Energy contract · Enumerated selection · Epigraph variable · Evaporation · Exchange

F — Filling · Finite horizon (acyclic mode) · Forebay level · Forward pass · FPHA · Future anticipated delivery

G — Generic constraint · GEVAZP

H — Head · HiGHS · Historical window · Hydro plant · Hydro unit group

I — Inflow · Inflow-lag seed · Innovation · In-sample sampling

L — Lead ( $K_i$ ) · Level-1 cut selection · Line · Load · Load factor · Lower bound · LP

M — Marginal cost (CMO) · Minimum generation

N — NEWAVE · Node · Node axis (output) · Non-controllable source (NCS)

O — ONS · Opening · Opening tree · Operating cost · Optimality gap · Outer approximation · Out-of-sample sampling

P — PAR(p) · Past anticipated commitments · Penalty slack · Policy graph · Post-study stage · PreFilling · Productivity · Pumping station

R — Reference volume · Relatively complete recourse · Reservoir · Right boundary / Terminal boundary FCF · Risk-averse · Risk-neutral · Run-of-river

S — Sample average approximation (SAA) · Sampling scheme · Scenario · Seasonal mean / std · Season function $\tau(t)$ · Security curve · Simplex method · SIN · Single-cut formulation · Spatial correlation · Spillage · Stage · Stage subproblem (stage LP) · State pinning (column bounds) · State variable · Storage · Stored energy (EARM) · Subsystem

T — Tailrace level · Terminal boundary cut · Thermal unit · Trajectory record · Transition (edge) · Transit seed · Travel time · Trial point · Turbined outflow

U — Upper bound · Upstream · Useful volume

W — Warm-start · Water value

Y — Yule-Walker equations

Power System

English	Portuguese	Definition
Block	Patamar	An intra-stage time period (e.g., peak, shoulder, off-peak) representing load-level variation within a stage. Novomodelo supports two block topologies: parallel (independent dispatches) and chronological (sequential, with carry-over storage). Brazilian practice names the three load blocks leve, média and pesada (light, medium, heavy load).
Bus	Barramento	A node in the electrical network where generation, demand, and transmission lines connect.
Line	Linha / Circuito	A transmission line or transformer connecting two buses, modelled with a directional capacity and an exchange cost.
Subsystem	Subsistema	A region of the interconnected grid (in Brazil: SE/CO, S, NE, N).
Exchange	Intercâmbio	Power transfer between buses across a transmission line.
Energy contract	-	A one-directional agreement to import power into the modelled system, or export power out of it, at a bus, with bounds on the contracted power and a price per MWh, a negative price being export revenue. An import injects power at its bus and an export withdraws it; a positive lower bound is a take-or-pay obligation. A contract carries no state and contributes nothing to the cuts. See Equipment-Specific Formulations §3.
Load	Carga / Demanda	Electrical power consumption at a bus, possibly varying per block via a load-factor table.
Deficit	Déficit	Unmet demand — load that cannot be served by available generation. Modelled as a slack variable with a high penalty cost.
Non-controllable source (NCS)	Fonte não despachável	A generating source, such as a wind farm, a solar plant or a small run-of-river hydro, whose available generation the solver takes as data: per stage and scenario, the installed capacity times an availability ratio from the source’s availability model, or the stage’s available generation when it has no such model, either way shaped across the blocks by a block factor. A curtailable source is dispatched anywhere between zero and its availability, with curtailment priced as a regularisation cost; a must-run source is pinned to its availability. Its generation is injected at one bus. See Equipment-Specific Formulations §6.
Generic constraint	-	A user-authored linear constraint over quantities of the stage LP, such as storages, flows, generations and deficits. Its shape — a floor, a cap, an equality or a two-sided band — follows from which bounds are given and, when both are, whether they coincide; it is active only at the stages, and optionally the blocks, for which bounds are given, and it may carry a slack priced per unit of violation, so that it relaxes instead of making the stage LP infeasible. See LP Formulation §10 and Generic Constraints.

Hydro Generation

English	Portuguese	Definition
Hydro plant	Usina hidrelétrica (UHE)	A hydroelectric generating station.
Hydro unit group	-	A group of a hydro plant’s turbines that carries its own bus and its own turbined-flow and generation bounds; every plant declares at least one. The plant keeps its water — storage, spillage, diversion and inflow are tracked once per plant — while the groups on one bus form a (hydro, bus) cell, with its own turbined flow and its own generation injected at that bus. See System Element Modeling Overview — Unit Groups and Bus-Partitioned Cells and LP Formulation §3.
Reservoir	Reservatório	Water storage volume behind a dam.
Storage	Armazenamento	Current water volume in a reservoir, in hm³. The primary state variable in SDDP.
Water value	-	The marginal value of stored water: the dual of a hydro’s water-balance row, in monetary units per hm³, the change in cost from one more hm³ of water in the reservoir. It is distinct from the storage cut coefficient, the reduced cost of the pinned incoming-storage column, which combines the water-balance dual with the duals of the other rows that read the incoming storage, such as the production-function and evaporation rows. See LP Formulation §4.
Inflow	Afluência / Vazão natural	Natural water flow arriving at a reservoir, modelled stochastically by a PAR(p) model (or 0-order seasonal sampling for the $p = 0$ degenerate case).
Turbined outflow	Vazão turbinada	Water passing through turbines to generate electricity.
Spillage	Vertimento	Water released from a reservoir without generating electricity.
Cascade	Cascata	Sequence of hydro plants along the same river, where downstream reservoirs receive turbined plus spilled water from upstream plants.
Downstream	Jusante	Direction of water flow; the plant that receives outflow from upstream.
Upstream	Montante	Direction against water flow; the plant whose outflow feeds a downstream plant.
Travel time	Tempo de viagem	The time a hydro’s release takes to reach its downstream plant along the main cascade arc. On an arc with a positive travel time, a share of each release reaches the downstream plant within the release stage and the rest in later stages; the volume still in transit is carried between stages as augmented state, pinned like storage and with its own cut coefficients. An absent or zero travel time is an instantaneous transfer, and diversion and pumping arcs carry none. See State Augmentation §6.
Productivity	Produtibilidade	Conversion factor from water flow (m³/s) to power (MW).
Run-of-river	Fio d’água	Hydro plant with no significant storage capacity; storage variable bounds collapse to a single point.
Diversion	Desvio	Water bypassed to a separate channel, not passing through turbines.
Pumping station	Usina elevatória	A station that pumps water from the reservoir of a source hydro into that of a destination hydro, drawing electrical power at its bus in proportion to the pumped flow. It carries no cost of its own: the power it draws enters its bus’s load balance, so pumping is priced by the marginal cost of energy there. See Equipment-Specific Formulations §4.
Evaporation	Evaporação	Water loss from the reservoir surface; can be negative under monthly-net conventions, requiring a bidirectional slack.
Forebay level	Nível de montante	Water level at the upstream face of the dam, a function of reservoir storage.
Tailrace level	Nível de jusante	Water level at the downstream channel below the dam, a function of total outflow.
Head	Queda	Height difference between forebay and tailrace levels, determining generation efficiency.
FPHA	Função de Produção Hidrelétrica Aproximada	Approximate Hydroelectric Production Function — piecewise-linear model of hydro generation as a function of storage, turbined flow, and spillage. See Hydro Production Function Models.
Reference volume	-	A plant’s reference operating storage at a stage, declared per production-model entry as an absolute volume or as a fraction of the useful volume, a default fraction applying when none is declared. It is the operating point at which the equivalent productivity of a plant on the FPHA model is derived. Through its forebay level, it also sets the downstream level at which the backwater-coupled tailrace curves of the plant immediately upstream are interpolated when that plant’s FPHA is fitted. It does not set the storage window over which the FPHA planes are fitted. See Hydro Production Function Models §5.
Dead volume	Volume morto	The portion of the reservoir below the plant’s physical minimum storage; reservoirs in commissioning are filled to dead volume before entering normal operation.
Filling	Enchimento de volume morto	The lifecycle phase of a hydro commissioned with a filling period, from its filling start stage up to, but not including, its entry stage, during which the reservoir impounds water toward its dead volume: turbined flow, generation and diversion are held at zero while spillage stays free, and a per-stage storage floor, relaxed by a penalised slack, rises at the plant’s minimum filling rate to reach the dead volume at the last filling stage. See LP Formulation — Lifecycle Phases.
PreFilling	-	The lifecycle phase of a hydro whose dam is not part of the dispatch at that stage: a filling hydro before its filling start stage, or, at every stage outside its commissioning window, a hydro without a filling period. Turbined flow, spillage and diversion are held at zero, storage is frozen at its incoming value, and the plant’s local inflow, the releases of its upstream plants, the flows diverted into it and the in-transit water maturing into it pass to the first downstream plant that is not PreFilling, or leave the system when there is none. See LP Formulation — Lifecycle Phases.
Useful volume	Volume útil	Storage above the plant’s physical minimum storage, up to its physical maximum. See Hydro Production Function Models §5.
Stored energy (EARM)	Energia armazenada (EARM)	Energy content of the stored useful volume: the volume above the physical minimum valued at the useful-range mean accumulated productivity, in MWh (or MW over the stage’s hours). See Hydro Production Function Models §5.
Security curve	Curva de segurança	A per-stage floor on stored energy, expressed as a fraction of the plant’s maximum stored energy. See Hydro Production Function Models §5.
Transit seed	Semente de trânsito	The simulation’s rolling release-window export (per scenario and hydro, dated windows with their release rates), which seeds the past releases of a chained study so that it can reconstruct in-transit state on its cascade arcs. See State Augmentation §6 and Simulation Output.

Thermal Generation

English	Portuguese	Definition
Thermal unit	Usina termelétrica (UTE)	A fossil-fuel or nuclear generating station.
Minimum generation	Geração mínima / Inflexibilidade	The minimum output of a thermal plant in service; Novomodelo enforces it as a hard lower bound on the plant’s generation in each block (a must-run floor), with no penalty slack, and the bound is zero outside the plant’s commissioning window.
Operating cost	Custo variável unitário (CVU)	Variable cost per MWh of generation.
Marginal cost (CMO)	Custo Marginal de Operação	Shadow price of the bus load-balance constraint — the cost of one additional MWh of demand at that bus.
Anticipated dispatch	Antecipação de despacho	A per-plant flag that splits the thermal decision into a commitment decided at an earlier stage and a delivered generation forced to match the commitment at the delivery stage. The lead is a number of stages or a physical lead in hours. Motivated by LNG (GNL) fuel-ordering lead times. See System Element Modeling Overview §4 and System Entity Files.
Lead ( $K_i$ )	Antecedência	The lead of an anticipated thermal is a number of stages, or a physical lead in hours resolved per delivery stage on the stage calendar (the decision stage contains the instant one lead before the delivery stage’s end; an instant on a stage boundary belongs to the earlier stage). $K_i$ is the plant’s ring depth — the most commitments it holds at once — equal to the lead for a stage-count lead. See System Entity Files.
Past anticipated commitments	Compromissos antecipados pré-horizonte	The commitments plant $i$ decided before the study begins; the in-study ones seed their ring slots at the first stage. Stored in the initial conditions as dated, delivery-anchored records (plant and window dates), each an externally-decided MW rate held constant over its window; a plant’s windows tile every delivery stage it decided before the study exactly — the leading in-study stages and any post-horizon (já-comandada in DECOMP) stages alike, never a stage the study itself decides (sunk: their fuel enters no objective; with a boundary loaded, a post-horizon window’s state contribution is folded into the boundary cuts’ intercepts). See Initial Condition Files.
Future anticipated delivery	Entrega antecipada futura	An in-study decided anticipated-thermal commitment delivered after the study horizon. Declared solely by a thermal-bound entry in the post-study stages input (plant, post-study stage, cost per MWh and MW range) — one per reached post-study stage — and charged on its decision column at the entry’s cost, discounted from the delivery stage, while the boundary prices the commitment it carries (zero terminal value when no boundary is loaded). The right-boundary counterpart to a Past anticipated commitment that delivers post-horizon. See System Element Modeling Overview §4 and Stage Files.

Stochastic Modelling

English	Portuguese	Definition
Scenario	Cenário	One realisation of uncertain quantities (inflows, load, non-controllable sources) along the planning horizon.
Stage	Estágio	A time period in the planning horizon — typically a month, week, or hour.
State variable	Variável de estado	Variables that link one stage to the next: storage volumes (always), AR-lag state (only for PAR(p) with $p \geq 1$ ), and, when the case declares them, the in-transit volumes of travel-time arcs and the commitment slots of anticipated thermals. See State Augmentation.
PAR(p)	PAR(p)	Periodic Autoregressive model of order $p$ for inflow generation. The order $p_m$ can vary by season $m$ ; $p_m = 0$ corresponds to white-noise (0-order) seasonal sampling.
0-order inflow	-	The degenerate $p = 0$ case of the PAR(p) model: $a_t = \mu_t + \sigma_t \varepsilon_t$ with $\varepsilon_t \sim \mathcal{N}(0,1)$ iid. No AR-lag state, no AR coefficients in the data file.
Innovation	Inovação	The independent standardised noise term $\varepsilon_t \sim \mathcal{N}(0,1)$ in the PAR(p) model, representing the unpredictable component after removing autoregressive structure.
AR lag	Defasagem	Past inflow values $a_{h, t-\ell}$ that enter the PAR(p) recursion for $p \geq 1$ . Stored as state variables in the LP and pinned to the previous stage’s realised inflow via column bounds.
Inflow-lag seed	Tendência hidrológica	The observed inflows before the study start that fill the AR lag state of the first study stages: the inflow history, overridden by the initial conditions’ recent observations where both cover a date. See Scenario Generation §4.5.
Yule-Walker equations	Equações de Yule-Walker	System of linear equations relating autoregressive coefficients to sample autocorrelations. Novomodelo uses the periodic Yule-Walker variant where the reference season shifts per row to handle multi-season covariance structure.
Spatial correlation	Correlação espacial	Cross-hydro covariance structure between innovations $\varepsilon_h$ . Applied via spectral factorisation $\varepsilon = C^{1/2} z$ with $C^{1/2} = U \Lambda^{1/2} U^{\top}$ , where $z \sim \mathcal{N}(0, I)$ .
Seasonal mean / std	Média / desvio sazonal	Per-season parameters $\mu_m, s_m$ , estimated from the inflow history or supplied in the inflow seasonal statistics input, which carries a mean and a standard deviation per hydro and stage. They are used both for sampling 0-order inflows and for converting standardised PAR coefficients to original units. See PAR(p) Inflow Model §3.2 and Scenario Files.
Node	Nó	A point in the policy graph: a unique identifier, the stage it sits at, an optional pointer into that stage’s externally supplied realizations and an optional label. The implicit stage chain — the graph every prior chapter assumes — has exactly one unnamed node per stage; declaring nodes explicitly lets a stage hold several. See Policy Graphs and Policy Graphs — Implementation in Novomodelo.
Node axis (output)	Nó (eixo de saída)	The node column of every simulation entity output and of the paths output: the visited policy-graph node’s identifier — the declared node identifier when the study declares nodes, otherwise the 0-based stage position (not the declared stage identifier). Prefixes each output row alongside the scenario and stage columns. See Policy Graphs and Simulation Output.
Transition (edge)	Transição (aresta)	A directed edge in the policy graph: from a source node to a target node exactly one stage later, weighted by a transition probability. Distinct from a node’s own within-node opening set — an edge weight says nothing about which realization is drawn, and an opening says nothing about which child node is visited next. See Policy Graphs.
Opening	Abertura	A pre-generated noise vector $\varepsilon$ used to evaluate the backward pass at one branch of the scenario tree.
Opening tree	Árvore de aberturas	The fixed set of pre-generated noise vectors used in the backward pass; generated once before training begins and reused across all iterations. Each stage — or, under an explicit node graph, each node — carries a configured number of openings. See Scenario Generation, Policy Graphs and Stage Files.
Sample average approximation (SAA)	Amostragem aleatória simples (AAS)	The replacement of each stage’s noise distribution by a finite, equiprobable sample of noise vectors: the opening tree, on which the backward pass builds its cuts, so that the policy is trained on the sampled problem. The default sampling method draws the openings by plain Monte Carlo: independent standard-normal components, which the spatial-correlation factor then correlates. See Scenario Generation §2.3a.
Sampling scheme	-	The rule that selects the noise a forward pass uses at each stage, chosen per stochastic class (inflow, load, non-controllable sources): in-sample draws an opening of the opening tree, out-of-sample draws fresh noise from the applied model, external reads supplied scenarios, and historical replays the inflow record, for the inflow class only. It is independent of the sampling method that populates the opening tree; the backward pass always evaluates the opening tree. See Scenario Generation §3.
In-sample sampling	Amostragem in-sample	Forward-pass scheme that draws trajectories from the same opening tree the backward pass uses. Default in Novomodelo.
Out-of-sample sampling	Amostragem out-of-sample	Forward-pass scheme that draws fresh noise from the applied stochastic model of each class that selects it (inflow, load, non-controllable sources) for each iteration and trajectory, from a seed independent of the opening tree’s, applying each stage’s sampling method across the iteration’s trajectories (plain Monte Carlo at selective and historical-residual stages); its forward costs estimate the policy’s expected cost under the model rather than under the opening tree.
Historical window	-	A year of the inflow record, read as the year in which the first study stage’s season occurs, whose record holds an observation of every hydro for the season occurrence of every study stage. Historical replay and historical-residual openings draw from the pool of such years. See Scenario Generation.
Load factor	-	The per-block multiplier that turns a bus’s stage-level load into the load of each block: the block’s load is the stage load times the bus’s factor for that stage and block, a block without a given factor carrying the stage load unchanged. See Scenario Generation §5.2.

SDDP Algorithm

English	Portuguese	Definition
Stage subproblem (stage LP)	Subproblema de estágio	The linear programme solved at one stage given an incoming state and a scenario realisation. Carries the cuts accumulated for that stage as constraints on the future-cost variable $\theta$ .
Bellman recursion	Recursão de Bellman	Recursive equation $V_t(x_{t-1}) = \mathbb{E}_{\omega_t}\bigl[\min_{x_t, u_t}\, c_t(x_t, u_t) + d_{t \to t+1} \cdot V_{t+1}(x_t)\bigr]$ that defines the cost-to-go functions and underlies the SDDP backward pass.
Cost-to-go function	Função de custo futuro (FCF)	Expected cost from the current stage to the end of the horizon, as a function of the incoming state. Convex and piecewise-linear under LP subproblems.
Cut (Benders cut)	Corte (de Benders)	A linear inequality $\theta \geq \beta_0 + \beta^{\top} x$ providing a lower bound on the future-cost function. Generated in the backward pass from the LP reduced costs of the pinned incoming-state columns at trial states.
Cut intercept	Intercepto	The scalar $\beta_0$ in a cut, anchoring the hyperplane vertically.
Cut slope (cut coefficient)	Coeficiente do corte	The vector $\beta$ in a cut, giving the partial derivative of the future-cost function with respect to each state variable. Negative for storage (more storage means lower cost) under the methodology’s sign convention.
State pinning (column bounds)	Fixação de estado	Binding an incoming-state coordinate to the trial value $\hat{x}_{t-1}$ by setting equal lower/upper bounds on its dedicated LP column. The column’s reduced cost is the cut coefficient for that state directly, with no preprocessing. It is KKT-equivalent to an equality fixing row and adds no row to the stage LP.
Trial point	Ponto amostral	The state visited during a forward pass, used as the anchor where the backward pass evaluates per-opening LPs and aggregates a cut.
Forward pass	Passagem direta	Phase that simulates the policy by solving stage LPs sequentially under sampled scenarios, producing trial points and a statistical upper-bound estimate.
Backward pass	Passagem reversa	Phase that walks stages in reverse, evaluates all openings at each trial point, extracts duals, and aggregates one cut per stage.
Enumerated selection	Seleção enumerada	A backward/forward pass mode, chosen as the enumerated method of the scenario selection setting, that visits every node of the policy graph deterministically and reconstructs each node’s incoming state along the single path from its parent, instead of drawing a sampled trajectory. Requires a singleton within-node opening set at every node and in-degree 1 everywhere (a pure tree, no recombining joins); rejected at setup if the exact enumeration would overflow. Contrasted with sampled selection, which draws a configured number of trajectories, carries its own incoming state per trajectory and is not subject to either restriction. See Policy Graphs §6 and Configuration.
Single-cut formulation	Formulação de corte único	Aggregation scheme that produces one cut per (stage, trial point) by averaging per-opening cuts with probability weights. Novomodelo’s default. Contrasted with the multi-cut formulation (one cut per opening), which Novomodelo does not implement.
Outer approximation	Aproximação exterior	The piecewise-linear lower bound on the future-cost function constructed from accumulated Benders cuts. The primary output of SDDP training.
Cut pool	Conjunto de cortes	Collection of all cuts at a given stage. Append-only across iterations within one training run; cuts occupy stable, deterministic slot indices and are never deleted — only deactivated and later reactivated, which keeps the cut order reproducible.
Level-1 cut selection	-	Cut-selection strategy that, at every visited trial point, keeps each cut within a tie tolerance of the per-state maximum cut value (evaluated over all populated cuts, active and inactive) and deactivates cuts that fall short at every visited state. One of a value-based family that also includes LML1 (keeps only the oldest near-maximum cut per state) and Domination (uses a domination tolerance band); all three share one kernel and reactivate cuts symmetrically. See Cut Management §7 and Configuration.
Dynamic cut selection (DCS)	Seleção dinâmica de cortes	Cut-management mode, chosen as the dynamic method of cut selection, that retains the full cut pool but loads only a small resident subset into each stage LP, growing it lazily until no omitted cut is violated or a round cap is reached — so the per-solve LP stays bounded as the pool grows, and a solve that stops on no violation returns the full-pool optimum when every pool cut is a candidate for addition. Mutually exclusive with the value-based pruning family (Level-1 / LML1 / Domination). See Cut Management and Configuration.
Lower bound	Limite inferior	The first stage’s risk-adjusted value over its openings with the current cuts; non-decreasing as cuts accumulate. See Upper Bound Evaluation.
Upper bound	Limite superior	A statistical estimate from a sampled forward-pass simulation (under any forward scheme), carrying genuine sampling error and a confidence interval; the exact enumerated bound over an exhaustively visited scenario tree or population, carrying zero sampling error: $\sum w \cdot c$ under expectation, the nested risk recursion over the tree under a CVaR uniform across stages; or a deterministic estimate from the reserved inner approximation. Under a stage-varying measure the enumerated value is the expected cost, which bounds no risk-averse value.
Census simulation	Simulação censitária	The post-training simulation variant whose scenario population is every leaf path of the policy graph, visited under enumerated selection (exhaustive, not sampled) rather than a Monte Carlo draw. Its cost statistic is a weighted mean and a weighted population variance over that population — no Bessel correction, no confidence interval. See Upper Bound Evaluation §4.5.
Optimality gap	Gap de otimalidade	The upper bound minus the lower bound; its percent form normalises it by the lower bound. The primary convergence diagnostic; see Stopping Rules.
Bound stalling	-	Stopping criterion that fires when the lower bound stops increasing across a configurable window of iterations.
Dual variable	Variável dual / Multiplicador	Shadow price from an LP solution; indicates the marginal value of a constraint right-hand side. The companion sensitivity for a variable is its reduced cost — for a column pinned at equal bounds, the reduced cost is the marginal value of that bound, which is how Novomodelo extracts state cut coefficients.
Epigraph variable	-	The auxiliary LP variable $\theta$ that lower-bounds the true cost-to-go function $V_{t+1}(x_t)$ ; named after the epigraph of a convex function.
Relatively complete recourse	Recurso relativamente completo	Property that every stage LP is feasible for any incoming state and scenario realisation. Novomodelo ensures this via penalty slack variables on every constraint that could otherwise be violated.
Penalty slack	-	A slack column, priced per unit of violation in the objective, that relaxes a constraint an extreme scenario can make impossible to meet, such as deficit on the load balance or a shortfall below a minimum outflow, so that every stage LP stays feasible. The penalty costs follow a priority ordering in which deficit outranks the operational-constraint slacks; the configured costs are checked against it when the case loads, with a warning when the ordering is not respected. See Penalty System §2.
Trajectory record	Trajetória	Data structure capturing one stage’s forward-pass result: primal solution, dual solution, stage cost, and end-of-stage state. Used for cut-coefficient extraction in the backward pass and for simulation output.

Boundary Conditions and Horizon Modes

English	Portuguese	Definition
Policy graph	Grafo de política	Directed graph of nodes and transitions defining an SDDP problem’s stage structure: a node sits at a stage and may carry its own realization pointer, and a transition is a directed, probability-weighted edge from a node to a node exactly one stage later. The implicit stage chain — one unnamed node per stage — is the default every prior chapter assumes; declaring nodes explicitly lets a stage hold several. Novomodelo’s policy graph is finite-horizon only (acyclic); a cyclic policy graph is a reserved target design, rejected at load. See Policy Graphs.
Finite horizon (acyclic mode)	Horizonte finito	Policy graph with a single terminal stage. Its terminal cost-to-go is zero unless a fixed terminal function is imported from an upstream policy (a right boundary). Under the zero terminal value it is susceptible to the end-of-world effect: reservoirs are systematically emptied near the terminal stage. See Horizon Modes §1.
Directed acyclic graph (DAG)	-	A directed graph with no cycle. Novomodelo’s policy graph is a finite directed acyclic graph whose every edge advances exactly one stage, so every trajectory runs from a root to a leaf in a bounded number of stages; a recombining graph, in which a node has several parents, is acyclic without being a tree. A cyclic policy graph is a reserved design, rejected when the case loads. See Policy Graphs §4.
Cyclic mode	Modo cíclico	Reserved design. Policy-graph shape with a back-edge that would return from the last stage of a cycle to the first stage of the next repetition, for long-term studies where a finite terminal condition would distort the policy; a case declaring it is rejected at load. Replaces what older Brazilian literature calls “infinite horizon”. See Horizon Modes.
Season function $\tau(t)$	Função de estação	The position of stage $t$ within one cycle of length $M$ : $\tau(t) = (t-1) \bmod M + 1$ . In cyclic mode, cuts are pooled by season, not by absolute stage.
Cycle convergence inequality	-	The requirement $d_{\text{cycle}} = \prod_{t \in \text{cycle}} d_{t \to t+1} < 1$ for the cumulative discount around one cycle, ensuring the value function remains finite across infinite repetitions.
Discount factor	Fator de desconto	Multiplicative factor $d \in (0, 1]$ applied to the future-cost variable $\theta$ in the stage objective, reflecting the time value of future costs. Required (strictly less than 1) for cyclic-mode convergence.
Terminal boundary cut	-	A Benders cut of the fixed terminal function imported from one pool of an upstream policy, used to chain studies (e.g., a weekly study inheriting the cuts of the monthly policy’s pool priced at the weekly horizon end). See Post-Study Boundary & Chained Studies.
Right boundary / Terminal boundary FCF	Fronteira direita (FCF terminal)	A DECOMP-style terminal future-cost function loaded from an external policy checkpoint named by the run configuration’s boundary setting (a checkpoint path and an optional strictness flag). The imported cuts form a fixed terminal function at the study’s terminal stage, in place of the zero terminal value; it prices every coordinate of the terminal state the source models, including held-to-terminal anticipated-commitment and in-transit state ( $\beta \cdot x$ ). The source instance’s own calendar is reconciled onto the current study’s calendar by dated, hour-weighted fan-out. See Post-Study Boundary & Chained Studies and Configuration.
Post-study stage	Estágio pós-estudo	A post-horizon calendar segment declared in the post-study stages input, carrying the per-thermal cost and bounds that are the sole surface for declaring an in-study-decided delivery beyond the horizon. A post-study stage is never dispatched: it gives a date past the horizon a calendar position and declares, per anticipated thermal, the capability and cost of a delivery there. It is declared with or without a loaded boundary. See Post-Study Boundary & Chained Studies and Stage Files.

Risk Measures

English	Portuguese	Definition
Coherent risk measure	Medida de risco coerente	A risk measure satisfying monotonicity, translation equivariance, positive homogeneity, and subadditivity. CVaR is the canonical example used in SDDP.
CVaR	CVaR	Conditional Value at Risk at tail fraction $\alpha$ — the expected cost in the worst $\alpha$ -fraction of scenarios. The basis for risk-averse aggregation weights in SDDP.
EAVaR	-	Expectation plus Average Value-at-Risk: the convex combination $(1-\lambda)\,\mathbb{E}[Z] + \lambda \cdot \text{CVaR}_\alpha[Z]$ used as Novomodelo’s parameterised risk measure. $\lambda = 0$ recovers risk-neutral; $\lambda = 1$ gives pure CVaR.
Risk-neutral	Neutro ao risco	Optimisation that minimises expected cost only. Probability weights at cut aggregation are uniform $p_\omega = 1/N_t$ .
Risk-averse	Averso ao risco	Optimisation that penalises high-cost tail scenarios. Cut aggregation reweights toward worse outcomes via the EAVaR formula.

Solver and LP

English	Portuguese	Definition
LP	Programa Linear	Linear Program — an optimisation problem with a linear objective and linear constraints.
Simplex method	Método Simplex	Algorithm for solving linear programs by traversing vertices of the feasible polytope. Novomodelo’s default warm-start strategy targets simplex bases.
Basis	Base	The set of basic variables defining a vertex of the LP feasible region; reused across iterations for warm-starting.
Warm-start	Partida a quente	Reusing a previous solution basis to accelerate the simplex method on a modified LP, including reconstructing a stored basis onto a churned cut pool by slot identity. See LP Warm-Start.
Cost scale	-	One positive factor, fixed per study, by which every objective coefficient except the future-cost variable’s is divided before the stage LPs are solved, to condition them numerically. It leaves the constraints and the optimal decisions unchanged; cuts are held in the scaled units, and objective values, duals and cost breakdowns are multiplied back by the factor when reported. See LP Layout and Scaling §2.1.
HiGHS	-	Open-source LP / MIP solver used as Novomodelo’s default backend.
CLP	-	Open-source LP solver from COIN-OR, available as an alternative compile-time backend.

Equivalent terms in other planning tools

The tables below map Novomodelo’s modelling concepts to their terms in NEWAVE and DECOMP, and the FPHA symbols of Hydro Production Function Models §2 to the practitioner notation of NEWAVE, DECOMP and DESSEM.

Concept equivalents.

Novomodelo concept	Term in NEWAVE/DECOMP	Note
Bus	Subsistema	A NEWAVE/DECOMP subsystem corresponds to one bus (bus granularity is user-defined); see Subsystem.
Block	Patamar	Same concept.
Must-run non-controllable source	Usinas não simuladas	A source whose generation equals its available generation in every block, with no curtailment; see System Element Modeling Overview — Curtailable vs. Must-Run.
Anticipated thermal	GNL antecipado	A thermal plant whose generation is committed at a stage before its delivery stage; see Anticipated dispatch.
Water travel time	Tempo de viagem	A release on a plant’s main cascade arc reaches the downstream plant after the travel time and is carried as in-transit state until it arrives; see System Element Modeling Overview — Cascade Travel Time.
Dead-volume filling	Enchimento de volume morto	A plant in commissioning fills its reservoir up to its dead volume on a per-stage minimum-accumulation schedule, without generating, before it enters service; see System Element Modeling Overview — Dead-Volume Filling.
Dead volume	Volume morto	A hard storage floor, with two exceptions: a filling hydro has a soft floor once it enters service, and a hydro not in service has no dead-volume floor; see LP Formulation — Lifecycle Phases.
Stored-energy floor as a generic constraint	VminOP; Curva de segurança	A per-stage lower bound on the productivity-weighted storage of a set of plants, optionally stated as a fraction of their maximum stored energy; see Security curve and LP Formulation §10.
Stage-varying maximum-storage override	Volume de espera	The flood-control reserve is entered as a lower maximum storage at the stages it covers. The plant’s physical storage range, which the stored-energy quantities read, is unchanged; see Hydro Production Function Models §5.3.
Individual hydro plants	REE (Reservatório equivalente de energia)	Novomodelo represents every hydro plant individually, with its own storage and production model, and builds no equivalent energy reservoir; see System Element Modeling Overview §5.

FPHA notation. The symbols belong to the FPHA model.

Novomodelo	Practitioner notation	Meaning	Units
$\phi$	FPH	Hydro production function	MW
$v$	$V$	Reservoir storage	hm³
$q$	$Q$	Turbined flow	m³/s
$s$	$S$ / $Q_{ver}$	Spillage	m³/s
$g_h$	GH	Hydro generation	MW
$h_{fore}$	$h_{mon}$ (montante)	Forebay (upstream) level	m
$h_{tail}$	$h_{jus}$ (jusante)	Tailrace (downstream) level	m
$h_{net}$	$h_{liq}$ (líquida)	Net head	m
$h_{loss}$	$h_{PerdH}$ (perda hidráulica)	Hydraulic losses	m
$q_{out}$	$Q_{jus}$	Outflow below the plant ( $q + s$ )	m³/s

Brazilian Power-System Ecosystem

Term	Portuguese	Definition
ONS	Operador Nacional do Sistema Elétrico	The Brazilian grid operator.
CEPEL	Centro de Pesquisas de Energia Elétrica	R&D centre that develops the official NEWAVE / DECOMP / DESSEM dispatch programs.
SIN	Sistema Interligado Nacional	The Brazilian interconnected power system.
NEWAVE	-	CEPEL’s long-term hydrothermal-dispatch model (monthly stages, multi-year horizon).
DECOMP	-	CEPEL’s medium-term dispatch model (weekly resolution). The “DECOMP-style” scenario tree (deterministic trunk with branching at the last stage) takes its name from this model.
DESSEM	-	CEPEL’s short-term dispatch model (hourly / half-hourly, day-ahead).
GEVAZP	-	CEPEL’s synthetic scenario-generation tool for hydro inflows. It draws inflow residuals from a three-parameter lognormal distribution (CEPEL manual, section Distribuição Lognormal 3 parâmetros; see Bibliography). The same manual’s section Modelo Autorregressivo Periódico - Par(p) writes the autoregressive coefficients of the standardized series with the Greek letter phi and the monthly standard deviation with sigma; the PAR(p) Inflow Model calls these the standardized autoregressive coefficients and the seasonal sample standard deviation, and uses sigma for the residual standard deviation.
CCEE	Câmara de Comercialização de Energia Elétrica	Brazilian electricity-trading chamber.