FlexPowerHub™ offers a broad portfolio of forecasts for the balancing energy markets, currently for Austria and Germany. All our forecasts, depending on the forecast horizon, are calculated several times a day and actively monitored in case of disruptions.
In balancing energy, our focus is primarily on 3 target variables:
- Energy price
- Capacity price
- Retrieval rate
Specifically, the most relevant quantiles, for both prices and retrieval rates, are estimated here. These quantiles can also be output in the form of cumulative MW steps. In order to have a continuous estimate of the MOL for the optimization of the ongoing bid calculation, customer-specific MOL forecasts are derived from the quantile estimators. On the one hand, to prevent negative feedbacks of the actual bids on the forecasts, and on the other hand, to avoid identical bids. Our forecasting system offers all customers statistically exactly the same forecast quality.
As a data science company, we are constantly striving to improve and develop our forecasts. Our dynamic models are continuously reviewed, recalibrated and improved to provide our customers with the best possible forecasts for the energy markets.
For the new market design (PICASSO & MARI), our next step will be to convert and re-train the forecasting models for energy price and retrieval quote to 15-minute. In addition, a forecast for the capacity price is trained for the PRL.
In addition, a forecast for the power price is trained for the PRL.
Currently we offer 1 dynamic & 3 analytical forecasts for the target variables LP ( capacity price), AP (energy price) and AQ (retrieval quota). The available forecast lead times are 2DA (day-ahead), 1DA and ID 1HA – 6HA (intraday 1 hour – 6 hours). By the end of 2022, all forecasts for dynamic and analytical forecasts should be available. In addition, we plan to provide our own spot & intraday forecasts for our customers by the end of the year.
2DA – 2 days ahead
1DA – 1 day ahead
ID – intra-day (1HA – 6HA, hour ahead)
LP – Capacity-price
AP – Energy-price
AQ – Retrieval-quota
Q4 – End Q4 2022
Q3 – End Q3 2022
X – Available
While analytical forecasts are mostly univariate, our DeepLearning methods take into account many other aspects besides historical prices: Comodity prices, fundamental data, weather and calendar effects. A total of approximately 80 influencing factors are included.
We use both Machine and more specialized DeepLearning algorithms in multivariate modeling. All models and error metrics are verified or calculated by unseen (test) data. Currently, we predict the Capacity-price, Energy-price and Retrieval-quota for SRL+- & TRL+- for DE and AT.
Prerequisite for using FlexPowerHub™ forecasts is a basic license, which automatically entitles you to use all available analytical forecasts. They also have a modular “up-sell” option for using the FPH DeepLearning prediction modules.
They can also view all forecasts in your in the FPH Dashboard against actuals. There, MOLs as well as the development over time are visualized. In addition, the most relevant error measures (MAE, MAPE and RMSE) are calculated and displayed directly there in order to convince oneself of the current forecast quality.
Performance comparison to benchmark
We always compare our dynamic forecast with all our analytical forecasts. In this context, the most-recent (previous day’s value) is our “main” benchmark. With the new forecast, we beat the most-recent values for almost all product time slices & quantiles in the period 01.12.21 – 30.06.22, in some cases by up to 45% for individual market time units.
For the calculation, the MAE of the previous day’s benchmark was compared with that of the forecast. Positive values represent a percentage better performance of the forecast compared to the benchmark.
Would you like to find out more about our forecasts? – We will be happy to send you our product data sheet. We also issue a test data set if required. This allows you to benchmark our forecast quality yourself.