Calibration to Index

To ensure that our forecast provides realistic revenues, we run our battery dispatch model on historical prices to model what revenues we would expect to be achieved. We then compare that with the revenues actually achieved by batteries from our historical benchmarking data, and calculate what adjustments need to be made so that our model is aligned.

Imperfect Foresight

Our dispatch model gives the battery imperfect foresight of prices. More specifically, the battery can see the next hour of prices, but after that it is given visibility of smoothed prices. This means it knows the rough price shape but it can't dispatch perfectly against a volatile daily price shape. The model then iterates throughout the day, hour by hour, with the battery locking in charge/discharge decisions each time. We have tuned the level of smoothing based on historical battery performance, so that this smoothing gets us most of the way to a realistic revenue forecast.

Calibration

After this smoothing, we still want to make sure that battery revenues are as closely aligned as possible with actual historical revenues. We do this by simulating the revenues of each asset with this imperfect foresight, and comparing that with their actual revenue performance. To make sure this analysis isn't skewed by batteries that have had performance issues, or batteries with high contracted revenues (e.g. Victorian Big Battery), we calibrate based on a filtered list of assets that have been performing well. Below you can see a comparison of modelled vs historical revenues for each of these assets.

We then calculate the ratio between modelled and actual revenues to work out how we need to adjust modelled revenues for them to align with historical revenues. This gives us an overall calibration rate of 92%. This calibration rate is then used in the forecast, by multiplying the outputs of our imperfect foresight battery dispatch model by 92% to get a realistic battery revenue forecast.