Assistance System for AI-Based Monitoring and Prediction in Smart Grids

Abstract

The German energy transition confronts the operators of low-voltage grids with new challenges. Local energy producers or large consumers, like, e.g., solar panels, heat pumps, and e-mobility lead to unexpected grid behavior. Because current grids are only sparsely monitored, local unmonitored overloads or violations of the voltage range are possible. To overcome these difficulties a smart monitoring and prediction system is needed. The system must handle different data sources fast and efficiently, so the operators can react to local grid problems. This is solved by using a streaming service to aggregate the data efficiently. Then, the implemented data pipeline is used to train AI-based models to interpolate the unmeasured parts of the grid. These models consider both measured data and predictions, like load profiles and photovoltaic forecasts. Since the grid is not fully observed, a data generator that physically simulates detailed grid scenarios is used to generate large sets of training data. Finally, an interactive GUI is implemented to visualize the data monitoring and predictions in the context of the grid and thus strengthen the user’s trust in the system. The presented assistance system is developed in close cooperation with energy experts and grid operators.