Melbourne Cup Day: A Grid Frequency Analysis

Another year, and another “race that stops the nation.” While I’m not of the betting sort, and I didn’t really get into it as such, it’s almost a tradition that no matter where you are, at 3pm on the first Tuesday in November, your eyes are glued to a TV tuned to the broadcast, usually Channel 7.

This year, like several in the past, has sadly ended in tragedy for at least one horse. While I won’t be caught in a discussion about the ethics behind horse racing and animal “rights”, events like these have the potential to cause big problems.

I suppose anyone nowadays knows how vital the electricity grid is to our everyday lives. Few people, aside from engineers, actually understand how intricate and dynamic such a system is. In essence, it’s a fine balance between consumers and producers – any significant mismatches have the potential to cause large scale blackouts as there is no sizable energy storage in the grid whatsoever.

The Melbourne Cup, in my mind, is one of those peak-consumption days, generated by the tradition of standing around a TV. It should be possible to observe the state of our grid by watching the energy … while everyone else watches their TV.

How to Monitor the Grid?

Most people are probably aware of the concept of voltage, and might have seen their lights dim as a laser printer fires up or someone turns on the heater. At first glance, it would seem that a good way to gauge the demand on the grid would be to watch the voltage.

Unfortunately, this is not true. In fact, measuring the voltage at your socket really only tells you about the local state of your grid, probably up to your local padmount substation. That’s because, the dips in voltage you experience are the voltage losses due to impedance in the cables leading to your house – turning on your oven locally would result in a bigger change to the voltage than a neighbour turning on their oven (for example).

The other complicating factor is the mass of transformers in substations along the way. These are not fixed ratio transformers, as many people incorrectly believe, and are instead variable in steps. Mechanisms known as tap-changers change the ratio of the transformer slightly in order to compensate for voltage drops elsewhere to keep your supply relatively stable, or within limits.

Instead, we have to monitor frequency. The frequency is a good indicator of the balance of generation and consumption in the grid, as the grid is synchronized. This means the frequency measured from pretty much any socket in the grid will be the same. Increased load on generators causes them to slightly slow down, whereas reduced loads causes them to slightly spin up from the nominal frequency of 50Hz in Australia. This property is used for demand side management.

Running the Grid

In Australia, the grid is run by AEMO – the Australian Energy Market Operator. The NEM (National Electricity Market) serves the east coast of Australia as an interconnected system. In general, demand and supply is forecasted and scheduled a day in advance, with any immediate short-falls made up by regulation/peaking plants which respond to demand. AEMO publishes useful documents about the grid standards and services. One of the more interesting ones is the Frequency Operating Standards (excerpted below), from AEMO’s Power System Frequency and Time Deviation Monitoring Report – Reference Guide.

Under normal circumstances, with our nominal grid frequency of 50Hz, we can expect the frequency to sit between 49.85 to 50.15Hz about 99% of the time, and stray out to 49.75 to 50.25Hz at most. When a generation or load event occurs (tripping of generator, or lines) then the frequency window broadens to 49.5 to 50.5Hz. To the regular person, this seems fairly inconsequential, but even small deviations in grid frequency normally mean “big” things happening at the network scale. If the frequency drifts too far, bad things can happen – transformers won’t operate correctly (possibly saturate), motors may overheat or overspeed, clocks will fall out of sync.

In order to keep the grid in check, certain services are purchased in order to regulate the frequency. A list of them can be found in AEMO’s Frequency Control Ancillary Services document, excerpted below.

It seems the scale of the “regulation” is on a 6-second reaction, 60-second reaction and 5-minute reaction scale. As a result, it’s likely that large swings in loads will see a movement in grid frequency before these regulation plants start to correct for it.

It’s known that when severe grid disruptions occur, often the frequency of the grid swings wildly as the generators try to rebalance themselves with loads that are shedding as feeders drop-out in a cascading fashion. It pays to keep an eye on the frequency (or more importantly, the phase at certain points and hence the power flow itself), to make sure the grid is in sync and in balance.

Methodology

One of the things that inspired me to try this little experiment was the FNET World-Wide Frequency Map. The Power IT Lab of the Department of Electrical Engineering and Computer Science of The University of Tenessee, Knoxville have been deploying GPS-referenced frequency data recorders to better understand power grids and flows especially in grid-events.

Unfortunately, I don’t have anything near as fancy, and recognizing the need for stable and accurate frequency sources for measurement, I decided to go it anyway using my Tektronix PA1000 Power Analyzer with PWRVIEW software. The unit itself is specced for a frequency accuracy of 0.1% which is a pretty lousy (for this application) 0.05Hz, but I decided to have it powered on for a whole day before the first measurements were taken, and have it in a temperature stable air-conditioned room. Hopefully this will keep its quartz crystal reference stable enough, even if it’s not the most accurate.

To see the impact of the Melbourne Cup, I recorded the frequency data for the day before the cup, and the cup day itself, and plotted it. Of course, having one day of historical frequency data doesn’t tell us much, but should provide a little reference (aside from 0Hz deviation from 50Hz nominal).

Results

Thanks to AEMO (the Australian Energy Market Operator) and their permissive copyright permissions, it seems I can show you this particular graph, taken from the AEMO home page, which shows the 30-minute demand for the relevant period.

Despite the bottom saying it’s from 03/11/2014 to 06/11/2014, the graph was taken today on the 04/11/2014. Notice how today’s peak is taller than yesterday’s peak, and it seems to “obscure” the morning and afternoon peaks as normally. It seems that the grid was under control today, with no peaks in the pricing which indicates that the supply was well within the demand and the forecast was reasonably accurate.

30 Minute Demand and Price graph from AEMO, copyright 2014 AEMO.

Plotting starts at 2:30pm, with yesterday’s frequency trend in black, and the cup day in red. The plot shows frequency deviation in Hz from 50Hz, as opposed to absolute frequency. The graph scale shows 49.85Hz to 50.15Hz, the 99% time window for no-contingency event or load event.

Note that before 2:35pm, it seems everything is normal – grid is hovering back and forth across the 0-hz offset line, as it should be.

It seems that at 2:37pm, load started to grow all the way through to 2:43pm, where the load was growing above generation causing the frequency to mostly remain below the 50Hz mark but well within normal operating parameters.

At 2:43pm to 2:45pm, the generation has “caught up” to the demand – remember, this is a dynamic system and the frequency represents the instantaneous “balance” of the system rather than the absolute load. But then, people seem to be turning on their TVs some more, so we’re below the line mostly from 2:46pm to 2:54pm.

At 2:55pm, the local Zellweger off-peak signalling plant fires up and shoves a strong 1050Hz signal into the line which seems to cause spikes in the PA1000’s measurement, so ignore those spikes. But once the signalling stops, we can see at 2:58pm, again, there’s a short spike showing more demand.

The nominal time for the race is 3pm, but it’s always a minute or two late to begin. We can see there’s a nice spike downward just at the 3:01:30pm mark. Once the race gets underway, it seems that we’re back in the net positive again! This is probably the cumulative effect of the “response” to frequency deviation, causing a brief over-supply. Maybe it’s a control or load interaction, but it seems to “oscillate” somewhat for a while, on the positive side.

In the moments between the race and the awards ceremony, it’s slightly up and down. At the beginning of the awards ceremony, the frequency seems to go up strongly – and it did yesterday too, so maybe there’s an unrelated consumer that schedules their power usage to go off at around that time?

Continuing with the award ceremony, it seems that the generation is mostly ahead of the load. After that, it seems the curve goes up and down quite strongly – probably as the peak demand around 4pm is also present for the day, and unrelated to the cup. Towards 3:57pm, the largest downward excursion in grid frequency was experienced at -0.16Hz.

In all, when taking into account the normal grid frequency and the “claimed” measurement error window in the device, the grid was having a decently good day. Compared to yesterday, the frequency excursions seemed a bit wider, but that’s to be expected. It pretty much sat within 0.15Hz of the nominal. When I took the mean of the two sets of data, it came out to 49.999Hz for race day, and 50.000Hz for yesterday, so I suppose the PA1000 is probably a little more accurate than it is specified for. To be sure, I did check it with FNET’s map, and it was showing the same trends about 3-seconds earlier than the FNET map did (so it’s probably bang on).

Conclusion

It seems that our power grid here is trundling along just fine. The load imposed by lots of viewers of the Melbourne Cup is clearly visible in AEMO demand graphs, but because of sufficient generation capacity, the prices did not spike and the frequency was within their control band for “no contingency event or load event”, within the 99% band for almost all samples. With the equipment I have in hand, it doesn’t seem that I’ve been able to see anything unusual happening with the grid, although compared with yesterday, the frequency was a little less stable.

Good to know! But also, it seems that the load changes due to people turning on and off appliances, with the Melbourne Cup race near the normal daily “peak” of demand did cause some more frequency “oscillations” than normal as the generators tried their best to keep everything regulated.

It wasn’t anything drastic, and many other countries (especially developing ones) see much more drastic shifts in grid frequency as part of their daily operation, especially those with chronic energy shortages.

I wonder what the grid frequency graph would look like for a major thunderstorm?