FCS procurement; Price formation, which when efficient should lead to FCS prices not only reflecting the true cost of the service, but also its true value to the system; and Allocation of the cost of FCS.

This is the introduction. Nam mollis congue tortor, sit amet convallis tortor mollis eget. Fusce viverra ut magna eu sagittis. Vestibulum at ultrices sapien, at elementum urna. Nam a blandit leo, non lobortis quam. Aliquam feugiat turpis vitae tincidunt ultricies. Mauris ullamcorper pellentesque nisl, vel molestie lorem viverra at.

A more comprehensive exploration of the impacts of model simplifications could be conducted.

The more divergent a modelling scenario is from the current system conditions, the less relevant the results of re-dispatch simplification analysis. However, the proposed method offers an additional tool for assessing and improving our understanding of electricity systems as they currently operate. We think it is a straightforward argument that an improved understanding of the current electricity system will be helpful as we navigate the challenges associated with modelling the energy transition

Pump Storage

More specifically, as perhaps in the case of generator behaviour in the NEM, the results may highlight the limitations of system designs that obscure the detail of critical operational decision-making processes, degrading the ability of stakeholders to understand how the market will operate in potential future scenarios.

On the other hand, the use of re-dispatch simplification analysis may highlight that some components of electricity system, such as generator behavioural dynamics, cannot be well represented within models.

reasonable simplifications

For short-term modelling, such as the hour and day-head forward processes run by AEMO [45], precise calibration to the current system state might be desirable and achievable. Over the medium term, from months to perhaps serval years ahead, ideally, simplifications would provide consistent, if imperfect, results over many years of test data. Systems such as the NEM have 10-plus years of historical data available, making testing across a broad set of conditions feasible. For longer-term modelling, replicating the current system in detail is potentially less helpful because the causal structure of the energy system is likely to change significantly [30]. On the other hand, a detailed model of how the system functions with the current set of operational and market arrangements might be very useful for testing the strengths and weaknesses of these arrangements for managing changing conditions, such as increased variable renewables, energy storage and decreasing amounts of fossil fuel-based thermal generators.

Interestingly, in this case, error balancing does not appear to occur in the unit dispatch error metrics, see [7], although there is no evidence to suggest this will be true in general.

, see [8]

The new method is not without limitations, some simplifications can be tested with less uncertainty than others, and the method needs to be applied with consideration for the internal consistency of scenarios and how multiple simplifications may interact.

relatively good isolation

prices lower than it would be typically historically

typically historically

ernatively, a number of techniques have been developed that model the decision-making process of energy-constrained units such as hydro generators

While this may result in a better fit to historical data, it would not represent the underlying dynamics of hydro and battery generator behaviour as they respond to changing market and water availability conditions.

quantity price

However, because the opportunity cost of hydro and battery generation is not tied to a fuel cost but rather the option to energy and used it to generate power in the future when prices are higher,

, to statically bid at the cap contract strike price, 300 $/MW,

, for dispatch outcomes this was the contract bidding simplification, and for price outcomes this was SRMC bidding.

Although

model results

attribute an increased divergence between model results and historical data to a failure to improve the component representation, because we may have improved the representation but now the remaining errors within the model are less well-balanced

This is illustrative of the error balancing that can confound the testing of model simplifications with historical data. In this case, the use of re-dispatch simplification analysis has allowed us to track the error introduced by each subsequent simplification clearly showing that the improved performance is the result of error balancing.

dispatch outcome results shown in [7] pricing

imilarly to the dispatch outcome results, the pricing results suggest that simplifications that affect unit complete unavailability are likely to have a greater impact on outcomes than simplifications that affect unit partial unavailability assumptions.

However,

The difference in volume weighted average price between the No Partial Unavailability and No Unavailability scenarios is on average 1.6 times the difference between the SRMC and Contract Bidding and No Partial Unavailability scenarios

de-commitment or complete outages

affect unit complete unavailability are likely to have a greater impact on dispatch outcomes than simplifications that affect unit partial unavailability.

64.8

4.7

, across the three years,

Figure 7:

The potential impact of partial unavailability simplifications is estimated by the difference between the outcomes in SRMC and Contract Bidding and No Partial Unavailability scenarios. While the potential impact of complete unavailability simplifications is estimated by the difference between the No Partial Unavailability and No Unavailability scenarios.

That is to say, the historical bids for volume above the max availability are essentially meaningless data that the operators knew would not impact on dispatch outcomes.

wo of these scenarios are base cases, and the others progressively simplify out all partial unavailability, and then all unavailability.

an outage

Unit maximum availability might differ from the registered (nameplate) capacity for serval reasons. A unit might be operating with a maximum availability greater than its minimum stable operating level but less than its registered capacity, either because of a partial outage, to manage an energy constraint such as fuel availability, or because it is withholding capacity in an attempt to increase the energy price. For brevity, we will call this the partial unavailability case. The other case is when a unit has a maximum availability of zero or less than their minimum stable operating level either because of an outage or because they have de-committed (or are decommitting) for economic or maintenance reasons. We will call this the complete unavailability case.

in this section,

distribution

highlighting that the match with the historical distributions is much closer in 2020 than in 2018 or 2019. Although not shown, the distribution for the other scenarios in 2018, and 2019 are consistent with the 2020 results.

he high and low ends of the distribution, and a closer match occurs near the median

The removal of FCAS markets and ramp rates consistently causes the least deviation in yearly volume weighted average prices across 2018, 2019, and 2020. The removal of generic constraints and SRMC bidding assumptions cause similar deviations in yearly volume weighted average prices, with both showing significant variation between years. Contract bidding by hydro and battery generators causes the most deviation in yearly volume weighted average prices across all years.

peaking generators

removal of FCAS markets and ramp rates

Figure 5:

ompares

ted

. A moving average, calculated with a centred window of width 21, has been applied to the simulation results to highlight the underlying trends in price deviation from historical values.

,

,

In general,

black coal

highlighting that not all units or technology groups are affected uniformly,

Figure 4:

No FCAS markets - No ramp rates

only data for 2020 is plotted for better visual clarity.

Three measures are shown: unit capacity factor change distribution, normalised interquartile range in capacity factor change, and normalised mean absolute error (MAE) in dispatch targets.

The interquartile range for all units shown, normalised to the maximum value, with the scale on the right-hand side. Lastly, the mean absolute error for unit dispatch is shown, normalised to the maximum value, with the scale on the right-hand side.

only 2020 data is shown.

and aggregated

exclusion of generator ramp rates, FCAS market and generic constraints.

SRMC bidding

contract based bidding

Detailed descriptions of each simplification are presented in the methodology, [Tbl:simplification_descriptions?].

s that are not captured in the base case model.

the methodology

wieghted

logathrimic axis.

linear y-axis

The Mean Absolute Error (MAE) across all units and all intervals rounds down to 0.0 MW.

[2] shows the distribution of deviations of modelled unit dispatch targets from historical targets.

95 % of units had less than 1 MW of deviation from historical targets, and in 743 of the intervals, the maximum deviation of any unit was less than 1 MW

he purpose of the benchmarking is to demonstrate that the base dispatch model closely replicates the actual NEM dispatch procedure.

Technology aggregate dispatch for the 1000 random intervals tested from 2020.

better measure of the impact

small impact

he SRMC bidding simplification causes a larger change in dispatch, and this change increases when ramp rates and FCAS constraints are removed

by SRMC bidding

This phenomenon can be seen in [1].

Application of Ramp rate and FCAS simplifications

For this reason, all the simplification scenarios explored in the results, [sec:results?], include the removal of FCAS markets and ramp rates.

change

gisterd

Simplification definitions

described

De-commited, decommiting, or recommiting units bid parameters are not modified. De-commited units are those with no bid in volume or a maximum avialability parameter set to zero. De-commiting or re-commiting units are thermal units with an available capacity less than their minimum stable operating level, as indicated by the capacity offered at less than 0 $/MW.

sterd

are modified to reflect bidding

nd incentives

strike price

These contracts

Cap contracts are often used by peaking generators to manage spot market revenue risk

300 $/MW

$/MW

SRMC

CCGT

SRMC

Dispatch is conducted using a mixed integer linear program which maximises the value of spot market trade.

A summary of Nempy has been published in the Journal of Open Source Software [41], and more detailed documentation is available online

,

By comparing scenario results we aim to determine which system components have the greatest potential to change dispatch and pricing outcomes when their model representations are simplified.

chosen

The reason for adopting the per-interval approach is so the base case can be closely aligned with the real-world system and, the divergence of modelling outcomes from historical outcomes can be more precisely attributed to specific simplifications.

The results presented in [sec:benchmark?] indicate that the dispatch and pricing model implemented with Nempy for this work closely replicates the real-world procedure used by AEMO.

procedure used by AEMO

dynamic calculation of generic constraint right-hand side values based on unit and network SCADA. Instead, in this work, the right-hand side values calculated historically by AEMO are used.

Nempy implements the key market features of the NEM as described in [sec:background?]. Two di

A

market outcomes

simplifications are applied to the base case

costs of the supply-equals-demand and FCAS constraints to be used to determine marginal costs,

as part of the special ordered sets (type 2) formulation of interconnector losse

to

Unit commitment constraints can be submitted for units capable of reaching minimum stable operating levels within 30 minutes, so that commitment and decommitment decisions can be made in the dispatch procedure.

currently

availability

For each unit 10 price-quantity pair offers must be submitted by 1230 hours on the day ahead of dispatch, optionally in each of the energy and eight FCAS markets [45]. The quantity component of the offer can be re-submitted up until dispatch is run for the interval. A maximum availability in each market is also submitted, which acts as a cap on the quantity that can be provided.

if required

ramp rates

dispatch and spot market clearing

The rest of this paper is structured as follows. In Section 2, we provide background on the NEM. In Section 3, the method is described. In Section 4, we present and describe the results. Section 5, discusses the results. Finally, Section 6 presents our conclusions based on this work.

new evidence arising from this case study on which simplifications of the NEM are likely to have the most significant impact on model results.

Re-dispatch simplification analysis only partially addresses the problems created by confirmation holism because of two main limitations.

where additional model detail or effort to refine simplifications would be most effective

novel method, re-dispatch simplification analysis,

hat is to say, for example, market participants will have a small reaction to small changes in market conditions and a larger reaction to large changes in conditions

efficiently allocate complexity,

it

As the detailed model uses input data specific to individual historical dispatch intervals, such as unit initial operating conditions, it is not appropriate for forward-looking modelling studies. However, because of the data aggregation and decision-making role of centralised dispatch, the dispatch model still provides an opportunity to study the impact of a broad range of simplifications that might be applied in electricity system models more generally. For example, the impact of a simplified representation of market participant decision-making can be tested by modifying the bids submitted to the dispatch model, or a simplified network representation can be tested by modifying the network constraints.

using the precise system state at the start of the dispatch interval, rather than modelling intervals sequentially and using the results from one interval as the initial conditions for the next

compared to other approaches the current work still allows simplifications to be tested with fewer sources of confounding error, and a clearer understanding of how other simplifications contribute to model behaviour.

bid volume.

bids

availability

serval

However, identifying the components of model formulations where simplifications have the largest impact on modelling outcomes is a useful first step for prioritising which model components to study in detail.

While all their optimisation models show significant deviation from historical data, the results indicated that unit commitment constraints have a larger impact on model accuracy than reducing the detail of the network representation [22]. Unnewehr et al. tested the impact of network resolution on an optimisation model of the European energy system by comparing generator dispatch to historical data. Similarly, all their scenarios showed significant deviations from historical data. However, they found that increased network detail lowered the deviation from historical data for some generator types, including lignite, hard coal, and gas

electricity system models

pre-existing model

Further, the technique should be applicable to any electricity system that uses centralised dispatch.

. E

ree from significant errors so that deviations in results from historical values can be attributed with confidence to the simplifications being tested. Re

The benchmarking for our work is presented in [sec:benchmark?].

Re-dispatch simplification analysis is the use of a highly detailed model of a centralised dispatch procedure to test electricity system model simplifications.

re-dispatch

In restructured industries, these processes operate as markets and participants submit data that reflects the state of their assets, and operational decisions and preferences.

is often used

ggregate system data on network and unit operating states and costs, and use optimisation techniques

day-ahead, 30 to 5-minute-ahead basis,

Electricity systems are key parts of larger energy systems and are likely to become increasingly important as more energy end-use applications are electrified

energy systems

redible assumptions regarding how the various simplifications contribute to aggregate model error.

The results presented in [sec:availability_results_results?] demonstrate this phenomenon.

hypotheses that the change improved the component representation and that the original component formulation was not balancing out other errors within the model. Without further evidence, it is impossible to know which hypothesis to reject

we

we must assume

caused by

I

which has assisted in diagnosing this issue in energy system modelling.

aluable insight into the sensitivity of energy system models to formulation details.

their results showed similar average prices but larger deviations in time-resolved prices at the lower and higher ends of the distribution. They then tested 160 alternative implementations of the model finding that the selection of implementations for partial load efficiencies, unit technical constraints, time-coupling constraints, the network model, and temporal resolution are all important for balancing modelling accuracy and complexity

,

A small number of studies have combined benchmarking against historical data and model comparison to test how different simplifications affect accuracy.

data [30].

This approach can be taken further, through repeated testing models can be tuned to better recreate historical outcomes, for exampl

computation, model interpretation, and representational improvement

a role for methods that can assist in quantifying how well model simplifications represent the real-world energy system.

uncertainty

monte carlo

impacts of climate change on the energy system

short-term variability in long-term models

by

Recent reviews of energy system modelling tools argue for continued efforts to narrow the gap between model representation and the real-world energy system

inconsistency

However, achieving appropriate model detail is important

The scale and critical nature of modern energy systems largely preclude physical experimentation, as a result, modelling has become an important tool for generating knowledge

commercial decisions

In restructured energy industries

[1,2]

academic study

A similar set of issues are of concern to system planner and policy-makers who routinely use modelling to help understand emerging issues, and inform policy and planning decisions [9,10,11].

planner

As well as, more flexible energy resources and loads such as energy storage [5], electric vehicles [6], hydrogen electrolysers [7], and various forms of demand response [8].

Re-dispatch