At 18:06:47 on 16 November 2019 a power system experiment took place which demonstrated the frequency control response differences between traditional synchronous generation and inverter connected plant such as batteries, wind farms and solar farms.
SA was islanded from the rest of the NEM with a surplus of generation – as a result its frequency went up, and the generation in SA was required to rapidly ramp down in power output to prevent a system black out resulting from generators tripping on over speed.
What was interesting about this event was the different responses of the various equipment – the traditional gas turbine thermal plant, a solar farm (Tailem Bend), two wind farms (Lincoln Gap and Willogoleche) and two battery systems (Hornsdale and Dalrymple).
The draft power system event report prepared by AEMO (which is located here) presents high resolution graphs of the above mentioned devices and the virtual power plant.
I have omitted an analysis of the virtual power plant, not because it is not interesting, but because the data presented in the report did not appear to be high resolution.
When analysing several time series traces it is often useful to split each of the data series into components in order to simplify myriad traces into a fewer elements. These can be combined by summation to recreate original traces, but it is often useful to look at each component individually to get a deeper understanding of your data.
We have used a technique known as independent component analysis which splits the time series data into components which minimises their mutual information, i.e. from an information theory viewpoint the components are as different from each other as possible.
For this analysis we used 4 components to represent 12 high resolution traces (6 devices plus 6 frequency measurements). The table below shows the components and the summation coefficients to use to recreate the original 12 traces. To save space we have only shown the coefficients for one of the frequency traces because they are all very similar to each other.
The scales of the independent components are not shown because by definition they are normalized so that their mathematical norm is unity.
In the table above, red numbers indicate negative coefficients, blue numbers positive coefficients. We have arranged things so that the frequency trace has all negative coefficients. In order to effectively respond to the system frequency transient – ideally all the power trace components from the various installations should be positive – only the Hornsdale battery achieves this.
Because the independent component analysis minimizes mutual information – the signals can sometimes be given a physical interpretation. This is a dangerous practice because the process is a mathematical technique which does not directly reference the real world. Any interpretation is thus subjective and different people may place different interpretations on the squiggly lines they are looking at. For what it’s worth the following are my interpretations which are informed by my knowledge and experience of power systems.
IC1 appears to be sensitive to the step change in load response and damped oscillations from the rotating plant on the system.
IC2 appears to be picking up the “snap back” transient that has occurred on fault clearance of the original power system fault that caused SA to island.
IC3 appears to be sensitive to the fast-transients and is significant in the inverter connected equipment response. This could be due (for example) to the inverter connected plant returning to its market defined set point after the transient.
IC4 appears to be sensitive to the slower transients associated with all the various generation technologies. This could be (for example) the generation technology returning to its market defined set point after the transient.
Here is how the frequency signal is reconstructed from the Independent components.
Taking each installation in turn, we can see how responsive each has been to the frequency transient experienced by South Australia.
The thermal generation has reacted positively to the change in frequency for only one component, IC1 which we believe represents the bulk of the inertia response. It is negative for all the other components possibly because there was no allocation of Lower 6 second provision for the main thermal power stations on the FCAS market. Accordingly – except for the step change inertial response (which rotating plant cannot switch off) – the thermal generation contribution to frequency control appears to have been counterproductive.
The solar farm at Tailem bend had a positive response for IC1 and IC2, a negligible impact for IC4 and a negative impact for IC3. Our speculative view is that the control systems of Tailem Bend Solar farm have possibly been tuned to react too fast (this would be consistent with the requirements of the NER and current industry practices) – accordingly it tracked the “snap back” transient of IC2 but did not react correctly to the slower IC3 or IC4 transients.The actual power trace of Tailem Bend showed a more pronounced spike which the component reconstruction had difficulty representing.
The wind farms of Willogoleche and Lincoln Gap also had a positive response for IC1 and IC2 and negative responses for IC3 and IC4. Similarly, to the case for Tailem Bend Solar farm we speculate that this may have occurred because the control systems are tuned to respond too fast.
The Battery system response at Dalrymple was positive for IC1 and IC3, negligible for IC2 and negative for IC4. The Dalrymple BESS includes a form of ‘virtual synchronous generator’ functionality which possibly may have hampered its response for IC4 (which consists of the slower parts of the transient behaviour).
The Hornsdale battery which has a simple proportional controller (which we speculate would probably not be acceptable according to current industry practice) has a positive and accurate response across all independent components.
Subject to more detailed investigations which may result if more information becomes available. I think we can conclude the following:
– The inertia response of the synchronous machines can be replaced by invertor connected equipment which is properly configured. In general, the inverter connected plant is able to perform much more accurately than uncontrolled inertia.
– Inverter connected equipment which has been installed more recently than the Hornsdale battery do not necessarily have a better response. The results of the analysis indicate the response is likely worse, and we strongly suspect recent rule changes and industry practices to be the cause of this. Simplicity is often the best approach.
– The current design of the FCAS market which allow control system responses to be compromised by bidding behaviour is leading to reduced reliability for events such as discussed in this article. In my opinion control systems should be outside market oversight – they cost virtually nothing in the scheme of things so they should not be able to be detuned or switched off.
In contrast, reserve does cost money because it affects dispatch. The FCAS market should be reconfigured to reflect the actual physics that occurs so we have the most reliable system possible.
Brief Discussion of the effect of synchronous condensers
Reviewing the shape of each of the Independent components, what would be the response of a synchronous condenser to this event? The basis set of IC components would probably need to be modified to include damped oscillatory behaviour that would likely occur, but of the set of components shown above, IC2 (which represents a “snap back” transient due to the initial fault) would represent the behaviour of a Sync. Condenser the best.
In common with the traditional thermal generation – the effect on system frequency for this event would have been negative. The initial fault would have accelerated the speed of the Synchronous condensers and then when the fault was cleared the system would have struggled to slow down because of the extra inertia it was carrying.
Control is more important than inertia.
There is an everyday example which describes this. Imagine pushing around an empty shopping trolley around a supermarket aisle.
Now imagine pushing around a shopping trolley full to the brim of something heavy (say soft drink, water bottles or your favorite form of alcohol, or even bags of bird seed) .
Under which scenario would the trajectory and speed of the shopping trolley be easier to manage?
If you chose the lighter trolley, then I would agree with you.
Similarly, for this type of event –having the higher inertia on the SA system that Synchronous condensers would deliver is not an improvement – the analysis we have done indicates the opposite.
Bruce Miller is a power systems expert with Advisian. This article was first published on LinkedIn. Reproduced with permission.