Performance Test of a Neural Network based State Estimation for Low Voltage Grids with Weak Input Data

Conference: NEIS 2018 - Conference on Sustainable Energy Supply and Energy Storage Systems
09/20/2018 - 09/21/2018 at Hamburg, Deutschland

Proceedings: NEIS 2018

Pages: 6Language: englishTyp: PDF

Personal VDE Members are entitled to a 10% discount on this title

Weisenstein, Marco; Wellssow, Wolfram H. (Technical University of Kaiserslautern, Kaiserslautern, Germany)

This paper describes a performance test of a multi-layer perceptron neural network applied in the low voltage state esti-mation with weak input data, i.e. less measurements. Different from other neural network based state estimators, only measurements from the substation and from smart meters are used. In compliance with the current German law, only voltage magnitudes from smart meters are considered, which are regarded as non-private-consumption-related data. The tests focus on the evaluation of the estimation accuracy of the voltages and line currents at different input-settings. The results show significantly high estimation errors in some cases. The consideration of a time vector as input data provides no significant improvement. On the other hand, the results are not significantly worse if the measured values of the sub-station are not part of the input. Furthermore, a correlation between the magnitudes of the current and the estimation errors could be determined. Nevertheless, the results show a basic usability of neural networks used in low voltage state estimation systems but not with this configuration for direct use as state estimator.