AI-based forecasting and microforecasting of energy generation and consumption in decentralized structures
Brief description
The modeling of future grid loads is proving to be an extremely complex yet important field of research and application for science and practice in view of the large number of determinants. Based on a forecast that is as accurate as possible, grids should be designed and switched more appropriately, power plant capacities planned more efficiently and electricity storage technologies used more sensibly. Current grid management and the forecasting models used often reach their limits in the face of a rapidly changing energy market. Artificial neural networks (ANN) offer enormous potential for improving the quality of forecasting.
In contrast to conventional forecasting models, they are able to detect linear and non-linear correlations, including interaction effects of the variables fed in. The aim of the research project to be realized is to design and apply more powerful types of ANNs, such as Long Short-Term Memory (LSTM), as well as hybrid solutions consisting of ANNs and conventional forecasting methods. To this end, models are being developed using the example of a regional low-voltage grid, which can be adapted to the specific local conditions.
Such solutions are intended to increase the quality of forecasting and thus enable more efficient grid management. To realize the project, a modern IT infrastructure is also being developed that will allow all necessary data to be processed on the server side.