A Novel Data-Driven Method of Real-Time Transient Stability Assessment for AC/DC Hybrid Power Systems

To effectively tackle the operational challenges facing today’s bulk power systems with growing uncertainties, this paper presents a novel data-driven method for transient stability assessment method of the AC/DC hybrid power systems using auto-encoder-based algorithms for feature extraction and convolutional deep belief networks and Boltzmann machines for training accurate and robust models. First, a feature set is established from power grid operational states, and a stack-based noise reduction automatic encoder (SDAE) is used to extract key features that significantly affect the transient stability of the system. Next, transient stability assessment models are trained based on attention mechanisms, improved convolutional deep belief networks (CDBN), and multiple convolution-constrained Boltzmann machines to learn effective features from the input data adaptively. With such a design, the attention mechanism improves the model’s attention to instability samples, and the high-precision assessment of the systems’ transient stability is achieved. The experimental results conducted on the New England 39-bus system and the WECC 179-bus system show that the proposed method can accurately assess the transient stability of the system with multi-infeed DC by outperforming traditional supervised learning models considering various noise levels.