v3.3 release: changes since v3.2
8th Jan 2025
Happy new year!
The changes over the last quarter focus on how we model intraday revenues in the GB forecast. We have built an intraday price forecast and modified our dispatch model to use it, along with some other minor changes.
1. Forecast of intraday prices now available
We forecast uncertainties on demand, wind, solar, and plant outages, and use these to inform an intraday price forecast for RPD HH: the volume weighted average of all intraday confirmed trades of half-hour duration products. More information on how we do this can be found here.
The monthly and annual aggregate values of these are now available in our world view. If you subscribe to our power prices, you'll also see a new column for them there.
2. Dispatch model optimises against intraday prices in the future, rather than applying an uplift based on historic values
Using this forecast intraday price, we now bring it into our daily optimisation model.
The dispatch model now runs once for a day-ahead optimisation (using day-ahead wholesale prices and frequency response prices), and subsequently, the modelled battery gets repositioned in intraday markets. A rolling approach is applied, assuming 2 hours' worth of knowledge ahead of time of intraday prices. Any leftover flexibility the battery has is able to reposition against more favourable price spreads as the markets evolve to the delivery window. More detail on this is here.
The intraday revenues are now lower than v3.2, in line with reduced intraday volatility observed across 2024.
3. Commodity price update
Gas futures are up around 10% in the first 3 years of the forecast horizon. We have updated these as of 3rd Jan 2025.
The impact is higher wholesale prices and spreads across the first 3 years of the forecast, and so higher battery revenues.
4. Improvements to the modelling of the actions of the storage fleet
In line with continuous model improvements, we have improved the way the storage fleet (batteries, pumped hydro and other storage) responds to wholesale price. The result is higher spreads across the forecast horizon and slower retirements of low efficiency CCGTs.
5. Extensions to nuclear retirement dates
EDF has announced several changes to retirements of nuclear plants. These are incorporated into the nuclear capacity in v3.3.
Haysham 2 and Torness will now be retired in 2030 (it was 2028). Sizewell B is now set to retire in 2035 (it was 2033).
6. Changes to CCS buildout and battery storage buildout
- We no longer apply a flat 3GW cap to the capacity of battery storage that can be built in one year. Now, we apply a learning rate to this of 3% a year fro 2029 onwards, allowing more storage to be built in the 2030's.
- The hurdle rate for CCS now starts at 2% (previously, it was 0%). Once the fleet has reached 5GW, the hurdle rate increases by 0.2% for every additional GW of CCS capacity as the technology becomes more commercial. It's capped at 8%. The result is less CCS to 2035.
7. We've made the dispatch model run in 5 minutes
To improve the user experience in waiting for the results of custom runs to appear on the platform, these now run in around 5 minutes. Providing the results pass our automated quality assurance checks, they'll also now appear on the platform instantly.
So, kick off a custom run, go make a coffee, come back and your results will be ready and waiting!
8. We've tweaked how we calculate daily cycles
What Changed in v3.3?
Previously, we used a battery’s nameplate capacity to calculate cycles. As a battery degraded, its actual (usable) capacity would drop, but the capacity value in our calculation stayed the same—leading to lower reported cycles over time.
Now, we use the degraded capacity in our calculations. Even as the battery degrades, the number of daily cycles we report remains more consistent throughout its life.
Comparing v3.2 and v3.3
You may notice higher reported cycling for degraded batteries in v3.3. This isn’t due to a fundamental change in the underlying model, but rather this different approach to how we report cycling.