APIs are really fascinating. You enter a few data and immediately the API spits out a result, and that for really complex things like the forecast of a photovoltaic system. But how exactly does the program behind it work? What happens when we enter roof pitch, south orientation, nominal power and geo-coordinates of the PV forecast? To answer these and other questions, let’s take a look behind the scenes of SolarPredictionAPI.

Thanks to many incentive programs, PV systems are becoming more and more affordable. Thus, the installation is not only worthwhile for planet earth and our climate, but also for your personal financial goals in the long run. However, PV systems bring a natural challenge: the system’s power output is highly erratic, and a simple distinction between day = power and night = no power is usually too inaccurate to determine when to trigger power-intensive tasks such as charging an electric car.

In a sense, this volatility in electricity production is inherent and cannot be avoided; however, there are production prediction tools such as STROMDAO’s SolarPredictionAPI that can be used to predict a facility’s electricity production fairly accurately. By simply entering a few data, one can find out the production of the PV plant for every hour in a period of 4 days - and that easily without own measured values.

But how does this PV Energy forecast work?
The PV forecast is available to you as an API. The cool thing about APIs is that you don’t actually need to know what’s going on in the background. An API is often compared to a waiter in a restaurant. You do need to know how to communicate with the waiter, how to order, but you don’t need to know anything about what’s going on in the kitchen when the food is being cooked. However, just like in good restaurants, sometimes curiosity gets the better of you and you would love to know what is going on in the program behind the API. To satisfy this curiosity, we will now take a look behind the scenes at our PV forecast.

To forecast the production of the PV system you have to enter five values in the request:
wp - the nominal power (Wp) of the plant.
az - the south orientation of the plant, also called azimuth
deg - the roof pitch
at - the latitude of the geo-coordinates where the plant is located
lon - the longitude of the geo-coordinates where the plant is located.

After the request has been sent, a current weather forecast from various weather services is first called up from the specified geo-coordinates (parameters: lat, lon). This is already available in a database in the correct key figures, e.g. the global radiation in the hourly intervals. Thus, the important weather data can be retrieved quickly and do not have to be recalculated and compiled for each query.

Next, the correct forecast model for the PV system must be selected based on the parameters used. For example, a more west-facing PV plant needs a different prediction model than a completely south-facing plant (parameter: az). The different models are each based on an AI that has been trained by countless data from comparable plants. Once the best model for the particular plant has been selected, the PV forecast can be calculated. Depending on the model, this happens in different ways, but most often it is based on historical data from comparable plants combined with formulas for calculating electricity production. The results are then returned to the API, which outputs them in JSON format.

The AI’s various prediction models are constantly trained in the background and regularly updated to ensure accurate and up-to-date forecasts. This is also a major strength of the STROMDAO SolarPredictionAPI: the STROMDAO has access to a large amount of historical data from a wide variety of PV systems of different designs and locations. Due to this broad data base, the product can score with a very high accuracy. Due to a lot of expertise and experience with PV predictions, STROMDAO also has a very sophisticated selection process for the many models, which additionally contributes to the accuracy of the prediction. The SolarPredictionAPI is operated together with the GrünstromIndex in a green data center, which itself does not emit any greenhouse gases and whose waste heat is used for surrounding buildings.