Spatial-Temporal Dynamic Frequency Prediction Based on Integrating Model-Driven and Data-Driven

With the establishment of the network transmission pattern based on high-voltage direct current (HVDC) transmission and the massive grid connection of new energy, the risk of frequency instability or even frequency collapse in the system increases. The frequency after disturbance presents spatial-temporal dynamics. Any location in the grid with a frequency outside the allowed range may trigger frequency instability or even frequency collapse. Predicting the spatial temporal dynamic frequency response is essentially a problem of solving complex differential algebraic equations that describe the system, which traditional methods are very time-consuming and not generalizable and accurate enough to solve when online applications. In this paper, a method integrating long short-term memory network (LSTM) and average system frequency (ASF) to predict the spatial-temporal dynamic frequency is presented. The method uses the power system-related data before and after the disturbance for a short period of time as input features to predict the dynamic frequency of each generator node. Tests in the 39-bus system and 500-bus system show the generalizability and accuracy.