A Data-Knowledge-Hybrid-Driven Method for Modeling Reactive Power-Voltage Response Characteristics of Renewable Energy Sources

The problem of modeling reactive power-voltage response characteristics of renewable energy sources (RESs) is considered. A deep neural network (DNN)-based method is adopted to track time-variant response characteristics of renewable energy sources. In the DNN training, the response of the grid is considered in the DNN loss function. To adapt to time-varying parameters of reactive power-voltage response characteristics models, we propose an adaptive online modeling method to train typical deep learning (DL) models independently. Moreover, to explore temporal and spatial correlations of learning tasks, a knowledge graph (KG) is constructed to store, retrieve, and utilize pre-trained DL models. Benefiting from the fusion of KG and DL in adaptive online modeling, DL models are equipped with transferability among various conditions. Simulations on a modified IEEE39-bus system and a 2486-bus real system have demonstrated the effectiveness of the proposed approach.