Three headline observations, about the use of proxies for System Frequency in Frequency Control mechanisms

Like a growing number of stakeholders in the NEM, we’re gearing up for the transition into the new Frequency Performance Payments processes, the operational commencement of which is now just over 12 weeks away.

There have been a number of articles already shared here on WattClarity

… and there are a number of others still to come (including a couple today, such as this one).

In terms of prior articles, Linton has written a number, including the following:

• On 28th February ‘About System Frequency’
• Also on 28th February ‘Incentivising primary frequency response’ (outlining some of the changes)
• Also on 28th February ‘Only 14 weeks to go until FPP’ … now down to 12 weeks to go.
• On 4th March ‘Comparing Frequency Measures (Frequency, FI, FM)’ (note this one).
• On 6th March ‘Regulation prices and enablement – a historical review’ … in part because we can’t remember ever seeing such a 20+ year review of Regulation FCAS
• On 6th March ‘Overview of Regulation FCAS costs in 2024’ …. utilising the recently released GSD2024.

This process has been a learning exercise for me as well … and so I’d like to take this opportunity share the following three observations that I have gleaned through this process that perhaps did not resonate a strongly beforehand.

Observation #1 = We use a proxy for Frequency, and not Frequency itself, to apportion Regulation FCAS costs

Maybe I’m slow, but I did not strongly appreciate this until relatively recently.

In the old ‘Causer Pays’ process, and even in the ‘apportioning Regulation FCAS cost’ component of the new FPP arrangements it’s not the ‘raw’ frequency data that’s used to determine the apportionment of Regulation FCAS costs, but rather a proxy that’s derived from it:

• In the old (soon to be retired) ‘Causer Pays’ process it is the Frequency Indicator that is used
• In the new FPP process, it’s the Frequency Measure.

Linton noted more in this article … but the summary is:

• they are both calculated differently, drawing on frequency,
• but are not the same as frequency in each 4 second time snapshot.

How ‘not the same’ you might ask? …

Observation #2 = Frequency Indicator is a very poor proxy for Frequency

… well, we certainly asked this question.

So curiosity drove us to extract almost a year’s worth of 4-second data for each and we present the following comparisons – firstly for the mainland:

… and now for Tasmania:

The numbers staggered me, when I saw them.

This correlation analysis simply looks at each 4 second interval and divides them into four quadrants:

• • If frequency is above 50Hz:
    • Does the proxy for frequency suggest the need for lower services … if so, correlated
    • Does the proxy for frequency suggest the need for raise services … if so, anti-correlated
  • If the frequency is below 50Hz

    • Does the proxy for frequency suggest the need for lower services … if so, anti-correlated
    • Does the proxy for frequency suggest the need for raise services … if so, correlated

With the almost universal ‘red is bad, green is good’ colouring it helps to clearly just how amazingly woeful ‘is*’ the Frequency Indicator

* clearly not soon enough that is ‘was’

Observation #3 = Frequency Indicator is quite predictable … so that lends to pre-emptive behaviour

The third point I’d like to emphasise that jumped out at me in Linton’s article was how it’s relatively easy to predict, ahead of time, what the Frequency Indicator will be.

When we couple this with ‘Games that Self-Forecasters can play’, the picture becomes clearer.

Note that there is devil in the detail (e.g. both measures have in place provisions for excluding certain dispatch intervals in the apportionment of Regulation FCAS costs) … but these are ‘after the fact’ processes that don’t seem (to me) to exclude the upstream effect …

• that the bad correlation between the proxy for frequency and frequency itself will likely encourage behaviours that frequently worsen frequency, not improve it
• especially coupled with the fact that the proxy can be predicted ahead of time.