AI-Based State Estimation with Focus on Distribution Grids with few Measurements

Germany is planning a significant increase in renewable energy sources and the emergence of new consumer types by 2030. To manage this transition effectively, regulations have been established for Medium Voltage (MV) and Low Voltage (LV) Distribution System Operators (DSOs) to enhance grid monitoring and planning, aiming to prevent congestion caused by these new system participants. However, monitoring MV and LV grids presents challenges due to low measurement density, making real-time monitoring and planning essential yet complex.
Recent studies have demonstrated that artificial neural networks can improve visibility in LV and MV grids with low measurement density, showing promising results. Currently, we are testing state estimation methods using artificial neural networks on two realistic LV grids with artificially generated load time series. A viable approach for estimating future MV/LV grid states involves employing ensemble forecasts to generate probability distributions of grid state variables. This method accounts for uncertainties, particularly those arising from weather-dependent producers like photovoltaics (PV) and new consumers such as electric vehicles (EVs) and heat pumps.
Additionally, we present a novel tool for optimizing meter locations. While a transformer measurement may suffice at low grid utilization, it is advisable to measure additional cable distributors at higher utilizations to ensure effective monitoring and management of the grid.