We created a model that predicted the energy consumption of a power station one day in advance and could forecast the power generation potential of solar and wind stations.
Can you imagine spending a whole day without your phone, laptop or even without electricity at home? Most likely you can’t.
In today’s digital world, electricity is a vital factor in the country’s development. With our evolution, our dependence on electronic devices has also increased. Which, in turn, increased the demand for electricity use.
However, when it comes to energy production we should remember that energy cannot be stored in large quantities, therefore, ideally, energy generation and consumption should be balanced in real-time.
That is why the prediction of energy demand and the actual load on electrical grids is important. It helps companies make the right decisions regarding energy planning and generation.
Accurate forecasting of grid load is essential for proper planning and management of electrical power systems. Forecasting load is the basis for the energy price establishment as well for the demand side planning, management and energy storage scheduling.
Our client, one of the largest power supply alliances in Germany, located in the northeast of the country, often faced an unwanted situation where its existing distribution lines could not accommodate the required load during high demand or emergency situations. This occurrence is also known as grid congestion.
There was a need for a smart solution that would help the client avoid the power redispatching problem, altering the power generation and load pattern in order to change physical flows in the transmission system and relieve physical congestion.
Solution from actionable data
Load prediction is a complex problem and to develop a solution that accurately predicts grid load, it is important to understand what various factors influence energy demand.
We started our project by taking a look at conventional power station generation schedules as well as wind and solar power generation data. Also, we got information about the loads from up to 100 substations located in the northeast of Germany.
In addition to the data mentioned above, we analyzed the weather as an important factor in forecasting network load.
Climate plays an important role in forecasting energy demand, especially when the share of renewable energy has increased. Power generation from renewable energy sources like solar panels and wind turbines is highly dependent on weather conditions. Therefore, we analyze data related to temperature, humidity, wind speed and direction, cloud cover, and solar irradiance.
Based on all the data we had access to, we created a benchmarking model. The model was trained for each power substation individually. The client was able to predict the energy consumption of their power station one day in advance and could forecast the power generation potential of solar and wind stations. Based on this, the schedules of conventional power generation could be adjusted.
- Vertical Grid Load 24 hours ahead forecasting model with mean absolute percentage error (MAPE) of 92%
- Models for online Anomaly and Outlier detection/prediction online
- Proposals for data collection and additional data aggregation process
- Proposals for efficient application of open source Big Data management tools and applications
- Prophet, Decision Trees, PyTorch, scikit-learn
Are you interested in a similar project or do you want a different AI solution to improve your business operations? schedule a free call with our experts, we are always happy to chat!