Deep Learning-Based Real-Time Switching of Hybrid AC/DC Transmission Networks

Abstract

In this article, a new deep learning-based framework is developed to perform real-time switching of hybrid AC/DC transmission grids under the effect of dynamic line rating (DLR) constraint. The proposed deep learning model is designed to learn the topological patterns of buses/lines according to operational states, e.g., power injections and line impedances, in order to obtain the optimal switchings of transmission power grids. As the load and power generation vary with time, the network switching depends on both the current and the previous status of the load and the generation units. Thus, a deep learning-based time series model that integrates the gated recurrent unit (GRU) and the long short-term memory (LSTM) models is developed, which takes advantage of LSTM’s high accuracy and GRU’s high computational efficiency. The developed learning-based network switching framework is tested on a modified hybrid AC/DC IEEE 39-bus system, a modified hybrid AC/DC 118-bus test system, and a modified hybrid AC/DC 300-bus test system. Results show that the learning-based switching model can achieve both high accuracy and computational efficiency, which is suitable for real-time transmission network switching.