Distributed State Estimation in Digitized Low-Voltage Networks

This paper presents the first results of a research project which develops a new approach for the digitization of a low-voltage network. The approach uses artificial intelligence to reduce the requirements on measurement devices and parameterization. The paper focusses on the state estimation in low-voltage networks by means of artificial neural networks to estimate the grid state from exogenous inputs such as temperature and solar irradiance measurements, historical consumption and generation profiles, and measurements at the substation. Training of the neural network is completed in a central IT-system (cloud) where the necessary base data is available and can be advantageously complemented with data and operating experience from comparable low-voltage grids. For the actual operation, the trained neural network is deployed on the field devices where it uses real time exogenous inputs to estimate the states of the low-voltage network. The approach can be easily extended to sparse (or comprehensive) sensor placements below the substations. In this paper, we present the methodology of the state estimation approach and show first results for the estimation quality in the case when smart meter measurements of electrical loads, power injections and voltages at the consumers metering points are available but not in real-time. Moreover, we deliver insights into how well these loads can be estimated.