Knowledge Graph and Long Short-term Memory Network Fusion for Electric Power Sag Recognition

The electric power industry has advanced the requirements of energy conservation, environmental protection, and information technology, and the traditional power system cannot meet these requirements due to the growing global focus on environmental protection as well as the innovation and reform of energy management in various countries. Voltage sag recognition has received a lot of attention, however the current techniques still have major drawbacks. The classification and detection of voltage sag using a Knowledge Graph and long short-term memory network is proposed in this research (LSTM). First, a comprehensible voltage sag knowledge map for the regional power grid is built using data from the vast amounts of voltage sag data held in the power quality data center. The discretized knowledge in the knowledge graph is then transformed into low-dimensional real-valued vectors by using the TransH model to represent and understand the entities, relations, and relation planes of triples. The scoring function in TransH was then replaced with LSTM to improve the model’s capacity for learning. In order to direct the model learning process, the loss function is finally determined and employed as the model objective. Extensive studies on real-world datasets show that our framework is more effective than the baselines when compared to conventional machine learning methods and well-known neural network models.