+49 221 2612 167 info@energizedanalytics.com

Artificial Intelligence (AI) is a big topic these days. Many costumers ask us if EnBW is doing anything with Artificial Intelligence. Here is an example where AI is already operational within the EnBW.

We at SANDY developed a self-learning solution that predicts the consumption of a single household based its historic smart meter readings.

Our solution is part of the EnergyBase product, which intelligently manages the energy flow in households with e.g. a photovoltaic installation and battery storage. Together with a photovoltaic production forecast, the consumption forecast allows to quantify the energy that will be available to charge a battery or run other flexible consumers (heat pump or electric car). A good forecast allows increase the energy used within the household.

How to forecast the household consumption? If several hundred households were to be forecasted at the same time, the so-called HO-profiles could be used. However, the consumption of a single household differs significantly from a H0-profile. Furthermore, no two households are the same and thus the solution needs to learn each household individually. In addition, since the consumption pattern can change over time (e.g. kids move out, change in life-style) automatic adaption is needed. A third major requirement is robustness, since whatever the AI learned for a household, it will not undergo a final human supervision and result needs to be trusted.

The AI solution for forecasting the household consumption we derived fulfills all of these requirements.

The figures below demonstrate how the self-learning aspect of the solution works. Figure 1 shows the consumption (red) and the initial prediction (black) just after the installation. Since directly after installation, no historic data about is available, a H0 profile consumption pattern is returned. One can see that the agreement of consumption and prediction is poor. Figure 2 shows the predictions after 20 days. The AI already learned the main consumption pattern and prediction agrees fairly well with the actual consumption.

Um unsere Webseite für Sie optimal zu gestalten und fortlaufend verbessern zu können, verwenden wir Cookies. Durch die weitere Nutzung der Webseite stimmen Sie der Verwendung von Cookies zu. Weitere Informationen zu Cookies erhalten Sie in unserer Datenschutzerklärung

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.

Close