How are reserves dimensioned in Europe? 

Introduction 

The goal of the reserve dimensioning task is to find the minimum necessary volume of reserve capacity to ensure sufficient availability to cover the balancing needs in a Load Frequency Control (LFC) Area, following System Operation Guideline (SOGL) requirements. This text focuses its attention on the dimensioning of secondary and tertiary reserves, which in Europe are known as automatic/manual Frequency Restoration Reserves.

Under the regulatory frameworks established by the SOGL and the Electricity Balancing Guideline (EB GL), TSOs are moving away from traditional, static annual or quarterly reserve sizing toward dynamic, data-driven, and probabilistic approaches.

Which features can vary between approaches?

The estimation for Frequency Restoration Reserves (FRR) follows at least two of the following estimation approaches:

1. Deterministic (mandatory): The system must be robust enough to correct for the Reference Incident. According to SOGL, TSOs are required to ensure availability of at least as much balancing capacity as the largest supply unit (or importing interconnector) in the upward direction and as much as the largest demand unit (or exporting interconnector) in the downward direction.
2. Stochastic: The minimum volume is calculated based on the distribution of historical values of Area Control Error in open loop (ACEol). The ACEol is a measure of system imbalance before the activation of reserves. Using at least 1 year of historical records, TSOs should ensure enough FRR capacity to cover 99% of the imbalances.
3. Probabilistic: In this last approach, the stochastic estimation is enriched with the probability of outages in significant grid-connected equipment. Via simulated scenarios, the distribution of historical ACEol values is merged with outage probabilities of generators, loads, and interconnectors, and the probability of observing forecast errors. This method represents an improvement in reliability with respect to the stochastic one.

It is mandatory to ensure enough balancing capacity to cover expected imbalances within the LFC block for at least 99% of the time, based on the historical record. Therefore, either the stochastic or probabilistic approaches must be in place with historical ACEol records that comply with the SOGL.

Balancing products

While balancing products are standardised to a large extent in Europe, Nordic TSOs make a more granular classification of their primary reserves. Additionally, the products in Great Britain deviate from those in the Continental Europe Synchronous Area. The following table illustrates the quasi-equivalence between products.

Table 1. Balancing products and their equivalence across geographies in Europe.

Dimensioning dynamism

Timeframes are important in a highly dynamic system such as power networks. Reserve dimensioning and procurement used to be, for many TSOs, a yearly, quarterly, or monthly task. However, ignoring the information that becomes available in a shorter timeframe on the system conditions (e.g., weather conditions, electricity demand, grid loading, and outages) creates the risk of over/under-procuring capacity for long periods of time. 

If the reserve need is calculated as a single value over an entire year, extreme system conditions can be diluted by more frequent common conditions, resulting in underprocurement for extreme periods and overprocurement for common condition periods.

Figure 1. When imbalance is driven by predictable time-clustered system conditions, a static approach tends to both over- and under-dimension reserve needs. In the chart, static dimensioning covers 99% of the imbalances over the entire period, while dynamic dimensioning covers the same portion of imbalances but for each dimensioning period.

Therefore, the dimensioning period length and outlook are two dynamic design choices with a strong impact on balancing costs and reliability. The period length affects the clustering of similar imbalance data that will be used to ensure a level of reliability. The outlook determines the availability and accuracy of system condition forecasts. A more dynamic dimensioning approach has shorter dimensioning periods and outlooks and tends to use more available data than a more static approach.

Additional features 

There are some slight variations in the dimensioning approach used by some TSOs that can have an impact on their balancing capacity needs.

• Accounting for the variability of the Reference Incident. Accounting for planned maintenance and schedules of generators, loads, and interconnectors in the deterministic calculation prevents an overestimation of the FRR needs. For example, when a unit is expected to operate at most at 50% of its capacity because of weather or operational limitations, it is not necessary to dimension for the potential loss of this unit’s nominal capacity.
• Interconnector imports/exports. When the direction of interconnectors’ physical flows is certain, the potential loss of such interconnectors has an impact only on either the upward or the downward FRR capacity needs. Consequently, the needs in the unaffected direction can be reduced.
• Clustering of imbalance drivers. Using machine learning algorithms, the system conditions can be grouped into categories. Each category can also be associated with different imbalance drivers’ impacts. For example, periods with low renewable forecasts can have very different imbalance distributions than periods with high renewable forecasts.
• Consideration of non-contracted (free) bids. Forecasting the participation of non-contracted capacity in the energy activation market allows TSOs to safely reduce balancing capacity procurement. Grid reliability is maintained because these uncontracted volumes remain available to meet system needs.

N-SIDE has already collaborated with multiple TSOs, resulting in substantial savings in balancing costs. In our next blog, we delve into these specific projects and their impact.