State estimation in low-voltage grids by using artificial neural networks in consideration of optimal micro phasor measurement unit placement

Abstract:
At the low-voltage (LV) level there is mainly no measurement technology installed which is required for the implementation of a grid-compatible control of the charging of electric vehicles (EV) and feed-in of distributed energy resources (DER). This paper presents an artificial neural network (ANN) for state estimation (SE) in the LV grid, which uses the measuring data of micro-Phasor-Measurement-Units (µPMU). The ANN has been developed, optimized according to various hyperparameters, trained and validated on the basis of a 219 node LV grid combined with a synthetic load and generation profiles representing one month. It was investigated at which node and how many µPMUs have to be used in the 219 node grid for a sufficiently accurate voltage estimation. For this purpose, an algorithm for the optimal placement of the µPMUs in an LV-Grid has been developed to ensure an economic use of the measurement technology and the highest accuracy of the SE. The final results show that the developed ANN in combination with the placement algorithm can estimate the node voltage precisely enough without historical data for training and only a few µPMUs. In validation with 8 myPMUs used the mean absolute error (MAE) of the ANN estimating the node voltage was 0.771 V resulting in a mean absolute percentage error (MAPE) of 0.192 % and an root mean square error (RMSE) of 0.886 V. In another sce-nario, where an special case was assumed that half of all households were on holiday and also 8 myPMUs were used, the results were an accuracy of 0.870 V as MAE, a resulting MAPE of 0.217 % and an accuracy of 1.224 V as RMSE.