Nonintrusive Current Monitoring for Power Grids: An End-to-End Framework From Data Acquisition to Waveform Reconstruction

Current monitoring of ubiquitous nodes in power grids is becoming increasingly important as the global trend accelerates toward digitization. Nonintrusive current monitoring using passive wireless systems offers a promising approach to avoid issues, such as the complexities of deployment considering insulation requirements in wired solutions and the battery maintenance inconveniences encountered in the existing wireless solutions. However, the limited energy in passive wireless sensing nodes shortens the length of the collected sensing data, leading to challenges in reconstructing the physical information, especially when the nonlinearity of the sensors is unignorable. Here, in this article, we present an end-to-end framework on top of the developed passive wireless sensing system, spanning from sensing data acquisition to current waveform reconstruction. By exploiting the nonquadratic features of the sensor, the nonlinearity-caused lost polarity of the bipolar current-induced magnetic field in the sensing data is recovered using a bidirectional long short-term memory (LSTM) model, which further achieves the current waveform reconstruction as a nonlinear regression. Moreover, the synchronization of the sensing data and its ground truth current waveform are performed by phase estimation using polynomial fitting on the sensing data, addressing the model training challenges caused by the asynchrony between the two because of the passive wireless sensing mechanism. Experimental results show that the quoted errors of the current waveform reconstruction are within 0.68%, and the normalized correlation coefficients are above 0.99991. Our work provides a promising comprehensive solution to enable the evolution of the power grid to complete digitalization. It can be extended further to other physical quantity monitoring applications, empowering global digital intelligence.