An Enhanced TPA-LSTM Method for PMU Data Recovery and Prediction

The integrity of power system data is crucial for the operation and control of the grid. However,due to the integration of a large number of new energy sources and power electronic devices, phasor measurement units (PMU) suffer from all kinds of data losses. Therefore, this paper proposes a method that combines Enhanced Temporal Pattern Attention mechanism with Long Short-Term Memory (ETPA-LSTM). This method transforms the temporal pattern matrix of TPA to effectively extract features of different dimensions. It achieves high-precision prediction and recovery using only single-channel data from PMU. Case studies demonstrate that this method exhibits outstanding recovery accuracy under single-point and multi-point losses, as well as in scenarios of normal system operation and low-frequency oscillations. This method addresses the issue of data loss in complex systems and lays a solid foundation for the better application of PMU in power system.