Voltage-Based Heat Pump Recognition in Low Voltage Distribution Grids with Convolutional Neural Networks

Abstract

The increasing power generation by renewable energy plants in low voltage level leads to the need for further development of strategies for grid voltage stabilization. For this, there is the idea to gain information from the local grid environment of an inverter’s grid connection point by recognition and classification of electric loads based on the grid voltage to contribute to adaptive voltage control. This is solved by convolutional neural networks (CNNs) using a systematic training data generation, starting with power profiles and ending with scaled and noisy data. Hence, the proposed methodology achieves the goal without much simulation effort. Furthermore, it is shown that the CNNs can recognize a particular heat pump within realistic grid situations with an average accuracy of ca. 86%, while the accuracy is highly correlated to the distance of the measurement point to the transformer and the load to be recognized.