Line Parameter and Topology Estimation using a Deep Neural Network with Learnable Priors

The accurate topology and line parameters are crucial information for effective planning and controlling of the distribution systems. However, existing literature often assumes either full-scale deployment of micro-phasor measurement units (μPMUs) or the complete knowledge of the system topologies, which is unrealistic in practice. To address this, the author presents a deep learning-based approach for line parameter and topology estimation using historical measurements from μPMUs under partial observability. The proposed method incorporates a deep neural network (DNN) with a learnable regularizer, which takes into account the underlying system topology to achieve accurate and reliable line parameter and topology estimation. The proposed algorithm addresses the nonlinearity of power flows by unrolling the Gauss-Newton solver. This iterative process involves minimizing the losses and the learning regularization term alternatively. The proposed scheme’s performance is demonstrated on IEEE 33 test bus system, which showcases the algorithm’s efficacy.