Supervised Learning Approach for State Estimation in Distribution Systems with missing Input Data

The new global trend in the field of electrical energy supply is the application of artificial intelligence both for monitoring and controlling energy systems. High-performance power system state estimation is essential to ensure safe network operation. The approach presented here describes a method based on artificial neural networks to perform a state estimation for different switching states in distribution grids with missing input data. The approach is data-driven and based on synthetically generated data without consideration of historical measurements. With appropriate training, the method can predict grid variables, e.g., voltage magnitudes, voltage phase angles or line loadings, with high accuracy. Furthermore, the introduced procedure considers a high penetration of electric vehicles and photovoltaic systems. For measurement infrastructure planning, various configurations are presented to determine suitable measurement locations. The proposed concept is finally demonstrated using a synthetic reference grid.