Billions of data points
Data is essential when optimizing a battery. The more data you have, the more accurate your commercial strategy will be. But what do we mean by saying our trading models process billions of data points?
Photography works well as an analogy here. A camera with a high megapixel count will always produce a clearer image than its analog or low-megapixel counterparts. Standard imaging does the job – it captures your Berlin street art and New York City skyline, but it won’t show a supermassive black hole at the center of a galaxy 55 million light years from Earth. And continuously adding more data (or pixels in our example) will keep deblurring and upscaling the picture.
The accurate translation of all this data into executable trading decisions is as important as the data itself. Just like digital image processing will further refine our photo, a state-of-the-art optimization platform will increase the granularity of our trading models. Clearer visibility on market conditions enables strategic movement through a highly complex commercial environment.
How many data points does a successful trading strategy need?
The exact number of data points is difficult to determine, but it falls in the two-digit billion range and is steadily growing. In terms of data sources, we analyze the full order books of EPEX SPOT and Nord Pool, minute-by-minute forecasts for solar and wind, input from various grid operators as well as the information we collect from participating in the market. These data sets then get streamlined further for each relevant country, market, and revenue stack. Eventually, we process all gathered data with AI (artificial intelligence), ML (machine learning), and RL (reinforcement learning). In isolation, data points aren't of great actionable value, but collectively, they determine the marksmanship of a trading strategy.
Data as the blueprint for market navigation: A layered approach to BESS optimization
- Layer 1: Renewable forecasts
Next, we introduce data from renewable forecasts. By integrating expected outputs from solar, wind, and other clean sources, we refine our understanding of where surpluses and deficits may occur. This layer enhances our ability to predict the optimal times for charging and discharging. - Layer 2: Wholesale market data
We begin with the baseline – wholesale market prices. This initial data pool provides a rough outline of market conditions based on historical trends and future expectations driven by demand and generation profiles, setting the stage for our algorithms. - Additional layers: Gas prices, ancillary services rates, and more
Further granularity is added by incorporating data on gas prices, ancillary services, and other relevant indicators. Each additional layer sharpens our view, allowing us to develop a detailed schedule for battery operation over the next day.
Dynamic reoptimization
The result? A finely tuned operational schedule that maximizes the battery’s performance based on the available data. However, as time moves forward, conditions change—forecasts update, market prices shift, and the previously clear picture begins to blur again. At this point, our process kicks in once more: New data is layered in in real-time, and the schedule is reoptimized to keep pace with the evolving market landscape. Order books from EPEX SPOT and Nordpool provide input for the reoptimization – algorithms rerun every time there is a change in the order book, and models are updated in line with changes in the system.
The image below visualizes how data sharpens an optimization strategy through 3 stages:
- Blurry and sparse: The optimization strategy operates blindly
- Growing but noisy: The optimization strategy has direction but underperforms
- Rich, reliable, and high-quality: The optimization strategy delivers on all fronts, from maximized revenue and minimized cost to enhanced utilization and degradation-resilient operation.
Data in commercial BESS optimization
In the graphs below, you can see two backtests with diverging data input. Reference dashboard 1 shows a backtest of our cross-market BESS optimization over a one-week period for a 5 MW/10 MWh asset in Germany, using our usual data sets, including exact day-ahead (DA) auction outputs. In reference dashboard 2, the DA auction results are replaced with a constant number (average price of the previous week), meaning instead of a price curve with hourly data, we integrate a flat line. The outcome? Vague data leads to ineffective trading decisions by the algorithm, affecting commercial activities, battery health data, and ultimately revenue. Our inclusion of dynamic data points in just one aspect of the optimization brings a revenue increase of € 7,547.
Dashboard backtest 1
Dashboard backtest 2
The power industry runs on data
Data brings clarity. It transforms a vague, uncertain image into one that is detailed, actionable, and dynamic, ensuring that our energy systems remain agile and responsive in an ever-changing market. The bottom line is: High-quality data strengthens the trading strategy, and high-quality data interpretation achieves the most profitable outcome. By giving insight into the way we use data in our optimization, we are adding another piece to the big industry puzzle of transparency. Additional initiatives include giving customers 24/7 access to their trading dashboard and publishing certified portfolio performance with real revenues.
