Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 13 additions & 14 deletions docs/model/agents.md
Original file line number Diff line number Diff line change
Expand Up @@ -45,19 +45,17 @@ feasible options remain.

### Commodity prices used in appraisal

Investment appraisal uses three distinct price sets, all sourced from the previous MSY's dispatch
(or from the previous [ironing-out loop][framework-overview] iteration if enabled):
Investment appraisal uses two price sets calculated from the previous MSY's dispatch (or from the
previous [ironing-out loop][framework-overview] iteration if enabled):

- **Shadow prices** \\( \lambda\_{c,r,t} \\): used for activity coefficients in the mini dispatch
optimisation step of each appraisal (see [Mini Dispatch Optimisation](#mini-dispatch-optimisation)).
- **Market prices** \\( \pi\_{c,r,t} \\): used to calculate the investment metric (Cost Index or
SNAS) after dispatch.
- **Market prices** \\( \pi\_{c,r,t} \\): used for activity coefficients in the mini dispatch and
for the investment metric (Cost Index or SNAS).
- **Fallback prices** \\( \phi\_{c,r,t} \\): used to incentivise dispatch in the mini dispatch
optimisation when shadow prices alone are insufficient (see
optimisation when market prices alone are insufficient (see
[Mini Dispatch Optimisation](#mini-dispatch-optimisation)). Calculated using the strategy defined
by `fallback_pricing_strategy` in [`model.toml`][model-toml].

See [Commodity Prices][prices] for how these price sets are calculated.
See [Commodity Prices][prices] for how prices are calculated.

### Capacities of options

Expand Down Expand Up @@ -126,7 +124,7 @@ elsewhere.
This is captured by the activity coefficient \\( \alpha_t \\), which combines two components:

\\[
\mathrm{NetOperatingCost}_t = \mathrm{OperatingCost}(t) - \mathrm{RevenueFromFlows}(\lambda, t)
\mathrm{NetOperatingCost}_t = \mathrm{OperatingCost}(t) - \mathrm{RevenueFromFlows}(\pi, t)
\\]

\\[
Expand All @@ -138,12 +136,12 @@ This is captured by the activity coefficient \\( \alpha_t \\), which combines tw
\\]

where \\( \mathrm{RevenueFromFlows} \\) is the sum of all commodity flow revenues and costs (positive
for outputs, negative for inputs) valued at shadow prices, \\( \mathrm{OperatingCost} \\) is the
for outputs, negative for inputs) valued at market prices, \\( \mathrm{OperatingCost} \\) is the
variable operating cost plus levies and flow costs, \\( f_{c,\mathrm{primary}} \\) is the primary
output flow coefficient, and \\( \varepsilon \\) is a small positive constant added to ensure that
break-even assets are still dispatched.

**NetOperatingCost** is the net cost of running the asset for one unit of activity at shadow prices
**NetOperatingCost** is the net cost of running the asset for one unit of activity at market prices
— negative when the asset is profitable (revenues exceed costs).

**FallbackCost** is the cost of procuring one unit of activity's worth of primary output from an
Expand Down Expand Up @@ -188,7 +186,8 @@ dispatch optimisation, and are applied to each asset separately.

After the dispatch LP is solved, an investment metric is calculated for the option using **market
prices**. The metric is based on the option's marginal activity, while the fixed costs and market
costs are those of the option itself.
costs are those of the option itself. Fallback revenue and epsilon are not included in the metric
costs, though they can affect the activity profile used to calculate the metric.

#### Marginal activity

Expand Down Expand Up @@ -309,7 +308,7 @@ All per-flow costs (\\( \mathrm{cost}\_{\mathrm{input}} \\),
| AFC | £1,000/MW |
| Capacity | 100 MW |

##### Prices (both shadow and market prices are equal in this example)
##### Market prices

| Commodity | \\( t_0 \\) (Peak) | \\( t_1 \\) (Off-peak) |
| ----------- | -------------------- | ------------------------ |
Expand All @@ -319,7 +318,7 @@ All per-flow costs (\\( \mathrm{cost}\_{\mathrm{input}} \\),

#### Mini Dispatch Optimisation (identical for LCOX and NPV)

Activity coefficients use shadow prices:
Activity coefficients use market prices:

**\\( t_0 \\):**
\\[
Expand Down
9 changes: 2 additions & 7 deletions docs/model/dispatch_optimisation.md
Original file line number Diff line number Diff line change
Expand Up @@ -151,8 +151,7 @@ These constraints are defined in the optional `commodity_constraints.csv` file.

The dual values (shadow prices) of the commodity balance constraints represent the marginal cost of
satisfying an additional unit of demand for that commodity in region \\( r \\) during selection
\\( s \\). These shadow prices are critical outputs of the dispatch model and are used to seed and
guide the investment appraisal of [competitive agents][competitive-agents] in subsequent steps.
\\( s \\). These shadow prices are used in in partial system dispatch.

---

Expand Down Expand Up @@ -245,9 +244,7 @@ candidate assets are present:
\end{cases}
\\]

where \\( \epsilon \\) is the `commodity_balance_epsilon` parameter. The shadow prices from this
candidate dispatch run are then used to seed and guide the investment appraisal of
[competitive agents][competitive-agents] in subsequent steps.
where \\( \epsilon \\) is the `commodity_balance_epsilon` parameter.

---

Expand Down Expand Up @@ -293,5 +290,3 @@ caused by one or more constraints defined in `commodity_constraints.csv`.
If the rerun remains infeasible, commodity constraints are not identified as the cause. Since
commodity constraints are an experimental feature, this diagnosis should be treated as indicative
rather than definitive.

[competitive-agents]: ./agents.md#competitive-agents
5 changes: 2 additions & 3 deletions schemas/input/model.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -57,10 +57,9 @@ properties:
notes: |
The fallback price represents the cost of procuring the commodity of interest from an
alternative source rather than producing it with the asset being appraised. It provides an
additional incentive for an asset to dispatch even when shadow prices alone are insufficient.
additional incentive for an asset to dispatch even when market prices alone are insufficient.

Setting this to `unpriced` uses a fallback price of zero, reverting to the pure
shadow-price formulation.
Setting this to `unpriced` uses a fallback price of zero.
value_of_lost_load:
type: number
description: The cost applied to unmet demand
Expand Down
12 changes: 9 additions & 3 deletions src/simulation/investment/competitive.rs
Original file line number Diff line number Diff line change
Expand Up @@ -236,7 +236,7 @@ fn log_on_equal_appraisal_outputs(
}

/// Get the best assets for meeting demand for the given commodity
#[allow(clippy::too_many_arguments)]
#[allow(clippy::too_many_arguments, clippy::too_many_lines)]
pub fn select_best_assets(
model: &Model,
mut opt_assets: Vec<AssetRef>,
Expand Down Expand Up @@ -265,8 +265,14 @@ pub fn select_best_assets(
prepare_commissioned_assets_for_retention(&mut opt_assets);

// Calculate coefficients for all asset options according to the agent's objective
let coefficients =
calculate_coefficients_for_assets(model, objective_type, &opt_assets, prices, year);
let coefficients = calculate_coefficients_for_assets(
model,
objective_type,
&opt_assets,
&prices.market,
&prices.fallback,
year,
);

// Iteratively select the best asset until demand is met
let mut round = 0;
Expand Down
89 changes: 28 additions & 61 deletions src/simulation/investment/competitive/coefficients.rs
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,6 @@ use crate::agent::ObjectiveType;
use crate::asset::AssetRef;
use crate::model::Model;
use crate::simulation::PriceMap;
use crate::simulation::prices::Prices;
use crate::time_slice::{TimeSliceID, TimeSliceInfo};
use crate::units::{MoneyPerActivity, MoneyPerFlow};
use indexmap::IndexMap;
Expand All @@ -13,35 +12,35 @@ use std::sync::Arc;
/// Per-time-slice cost coefficients for an asset.
///
/// These coefficients are calculated according to the agent's `ObjectiveType` and are used by the
/// investment appraisal routines. They comprise the activity coefficients (revenue minus operating
/// cost, derived from shadow prices) used in the appraisal optimisation, together with the market
/// costs (derived from market prices).
/// investment appraisal routines. Both the activity coefficients used in the appraisal optimisation
/// and the costs used to calculate the appraisal metric are derived from market prices.
#[derive(Clone)]
pub struct ObjectiveCoefficients {
/// Cost per unit of activity in each time slice
/// Net revenue per unit of activity, including fallback revenue and epsilon, in each time slice
pub activity_coefficients: IndexMap<TimeSliceID, MoneyPerActivity>,
/// Market costs associated with asset for each time slice
pub market_costs: IndexMap<TimeSliceID, MoneyPerActivity>,
/// Costs per unit of activity for the appraisal metric, excluding fallback revenue and epsilon
pub metric_costs: IndexMap<TimeSliceID, MoneyPerActivity>,
}

/// Calculates cost coefficients for a set of assets for a given objective type.
///
/// Returns a map from each asset to its [`ObjectiveCoefficients`], which holds a per-time-slice
/// activity coefficient and market cost.
/// activity coefficient and metric cost.
///
/// Activity coefficients are revenue from flows (including the primary output) minus operating
/// cost, calculated using shadow prices. A small positive epsilon is added to each activity
/// coefficient so that assets with near-zero net value still appear in dispatch.
/// cost, calculated using market prices, plus the primary output's fallback revenue. A small positive
/// epsilon is added to each activity coefficient so that assets with near-zero net value still appear
/// in dispatch.
///
/// Market costs are calculated using market prices rather than shadow prices. For NPV they use the
/// same revenue-minus-operating-cost calculation as the activity coefficients. For LCOX the sign is
/// inverted (as the value represents a cost) and the primary output (commodity of interest) is
/// excluded.
/// Metric costs are operating cost minus flow revenue at market prices. NPV includes all flows,
/// whereas LCOX excludes revenue from the primary output (commodity of interest). Neither includes
/// fallback revenue or epsilon.
pub fn calculate_coefficients_for_assets(
model: &Model,
objective_type: ObjectiveType,
assets: &[AssetRef],
prices: &Prices,
market_prices: &PriceMap,
fallback_prices: &PriceMap,
year: u32,
) -> HashMap<AssetRef, Arc<ObjectiveCoefficients>> {
assets
Expand All @@ -51,7 +50,8 @@ pub fn calculate_coefficients_for_assets(
asset,
objective_type,
&model.time_slice_info,
prices,
market_prices,
fallback_prices,
year,
);
(asset.clone(), Arc::new(coefficient))
Expand All @@ -64,79 +64,46 @@ fn calculate_coefficients_for_asset(
asset: &AssetRef,
objective_type: ObjectiveType,
time_slice_info: &TimeSliceInfo,
prices: &Prices,
market_prices: &PriceMap,
fallback_prices: &PriceMap,
year: u32,
) -> ObjectiveCoefficients {
// Small constant added to each activity coefficient to ensure break-even/slightly negative
// assets are still dispatched
const EPSILON_ACTIVITY_COEFFICIENT: MoneyPerActivity = MoneyPerActivity(f64::EPSILON * 100.0);

// Activity coefficients
let mut activity_coefficients = IndexMap::new();
let mut market_costs = IndexMap::new();
let mut metric_costs = IndexMap::new();
let primary_output_flow = asset.primary_output().unwrap();
let asset_region = asset.region_id();
for time_slice in time_slice_info.iter_ids() {
// Get the operating cost of the asset. This includes the variable operating cost, levies and
// flow costs, but excludes costs/revenues from commodity consumption/production.
let operating_cost = asset.get_operating_cost(year, time_slice);
let net_operating_cost =
-calculate_asset_revenues(asset, operating_cost, time_slice, &prices.shadow);
let net_revenue = asset.get_revenue_from_flows(market_prices, time_slice) - operating_cost;

let fallback_cost = prices
.fallback
let fallback_revenue = fallback_prices
.get(&primary_output_flow.commodity.id, asset_region, time_slice)
.unwrap_or(MoneyPerFlow(0.0))
* primary_output_flow.coeff;

activity_coefficients.insert(
time_slice.clone(),
fallback_cost - net_operating_cost + EPSILON_ACTIVITY_COEFFICIENT,
fallback_revenue + net_revenue + EPSILON_ACTIVITY_COEFFICIENT,
);

let market_cost = match objective_type {
let metric_cost = match objective_type {
ObjectiveType::LevelisedCostOfX => {
calculate_asset_costs_for_lcox(asset, operating_cost, time_slice, &prices.market)
}
ObjectiveType::NetPresentValue => {
-calculate_asset_revenues(asset, operating_cost, time_slice, &prices.market)
operating_cost
- asset.get_revenue_from_flows_excluding_primary(market_prices, time_slice)
}
ObjectiveType::NetPresentValue => -net_revenue,
};
market_costs.insert(time_slice.clone(), market_cost);
metric_costs.insert(time_slice.clone(), metric_cost);
}

ObjectiveCoefficients {
activity_coefficients,
market_costs,
metric_costs,
}
}

/// Calculate the revenue from all flows minus operating cost
fn calculate_asset_revenues(
asset: &AssetRef,
operating_cost: MoneyPerActivity,
time_slice: &TimeSliceID,
prices: &PriceMap,
) -> MoneyPerActivity {
// Revenue from flows including the primary output
let revenue_from_flows = asset.get_revenue_from_flows(prices, time_slice);

// The activity coefficient is the revenue from flows minus the operating cost (net revenue)
revenue_from_flows - operating_cost
}

/// Calculate asset costs for LCOX objective.
///
/// Excludes revenues from the primary output (commodity of interest).
fn calculate_asset_costs_for_lcox(
asset: &AssetRef,
operating_cost: MoneyPerActivity,
time_slice: &TimeSliceID,
prices: &PriceMap,
) -> MoneyPerActivity {
// Revenue from flows excluding the primary output
let revenue_from_flows = asset.get_revenue_from_flows_excluding_primary(prices, time_slice);

// The activity coefficient is the operating cost minus the revenue from non-primary flows
operating_cost - revenue_from_flows
}
8 changes: 4 additions & 4 deletions src/simulation/investment/competitive/metrics.rs
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,7 @@ pub struct AppraisalOutput {
pub unmet_demand: DemandMap,
/// The comparison metric to compare investment decisions
pub metric: Option<Box<dyn MetricTrait>>,
/// Activity coefficients and market costs used in the appraisal
/// Activity coefficients and metric costs used in the appraisal
pub coefficients: Arc<ObjectiveCoefficients>,
/// Full solution to the optimisation
pub solution: IndexMap<(AssetRef, TimeSliceID), Activity>,
Expand Down Expand Up @@ -233,7 +233,7 @@ fn calculate_lcox(
asset.total_capacity(),
annual_fixed_cost(asset),
&results.asset_activity,
&coefficients[asset].market_costs,
&coefficients[asset].metric_costs,
);

Ok(AppraisalOutput::new(
Expand Down Expand Up @@ -278,7 +278,7 @@ fn calculate_npv(
asset.total_capacity(),
annual_fixed_cost,
&results.asset_activity,
&coefficients[asset].market_costs,
&coefficients[asset].metric_costs,
);

Ok(AppraisalOutput::new(
Expand Down Expand Up @@ -465,7 +465,7 @@ mod tests {
fn objective_coeffs() -> Arc<ObjectiveCoefficients> {
Arc::new(ObjectiveCoefficients {
activity_coefficients: IndexMap::new(),
market_costs: IndexMap::new(),
metric_costs: IndexMap::new(),
})
}

Expand Down
10 changes: 3 additions & 7 deletions src/simulation/investment/competitive/optimisation.rs
Original file line number Diff line number Diff line change
Expand Up @@ -219,20 +219,16 @@ pub fn perform_optimisation(
// something we should consider.
//
// A second scenario is where there are two different (non-equivalent) assets with identical
// marginal cost (despite having different VOC and/or flows), and where moving activity between
// marginal cost, and where moving activity between
// the assets isn't prevented by constraints (e.g. activity constraints or utilisation peak
// constraints). In this case there will be two identical corner solutions, and adding a tranche
// to asset A may inadvertently cause it to switch to the equally good solution where asset B
// wins, hence resulting in a decrease in activity (the opposite can also happen where adding a
// tranche to asset B switches it to a solution where asset B wins, resulting in artificially
// inflated marginal activity). For example, in the simple example for 2021 investments RELCHP
// and RGASPB have identical marginal cost in autumn.night and autumn.evening. This is a bigger
// problem, going back to the issues of degeneracy and non-deterministic results, and something
// we should try to understand better and address at the dispatch stage (which ultimately sets
// the shadow prices that determine marginal costs). In the case of investment appraisal, this
// inflated marginal activity). In the case of investment appraisal, this
// can artificially decrease/increase marginal activities resulting in nonsensical LCOX. This is
// a problem. It may help to seed each model with the previous solution, to reduce the risk of
// flip-flopping, but the underlying issue also needs to be addressed.
// flip-flopping.

Ok(ResultsMap {
asset_activity: marginal_activity,
Expand Down
4 changes: 2 additions & 2 deletions src/simulation/investment/competitive/output.rs
Original file line number Diff line number Diff line change
Expand Up @@ -121,15 +121,15 @@ mod tests {
#[fixture]
fn appraisal_output(asset: Asset, time_slice: TimeSliceID) -> AppraisalOutput {
let activity_coefficients = indexmap! { time_slice.clone() => MoneyPerActivity(0.5) };
let market_costs = indexmap! { time_slice.clone() => MoneyPerActivity(0.4) };
let metric_costs = indexmap! { time_slice.clone() => MoneyPerActivity(0.4) };
let activity = indexmap! { time_slice.clone() => Activity(10.0) };
let unmet_demand =
indexmap! { TimeSliceSelection::Single(time_slice.clone()) => Flow(5.0) };
AppraisalOutput {
asset: AssetRef::from(asset),
coefficients: Arc::new(ObjectiveCoefficients {
activity_coefficients,
market_costs,
metric_costs,
}),
activity,
unmet_demand,
Expand Down
Loading