School of Electrical Engineering, Xi’an Jiaotong University, Xi’an710049, China

The accurate and effective transient stability assessment for power systems is of great significance for the safe and stable operation of power systems. At present, transient stability assessment methods based on deep learning are faced with problems such as difficulty in time-series feature space representation and serious imbalance of sample categories, which affect the reliability of assessment results. In order to make up for the shortcomings of existing studies, a transient stability assessment method for power systems based on the imbalanced sample enhancement of denoising diffusion probabilistic model (DDPM) is proposed. First, an improved HSV colour model is constructed to process the high-dimensional data in two-dimensional image, so as to visually represent the high-dimensional data and facilitate subsequent training. Then, based on DDPM algorithm, the imbalanced unstable sample space is characterized and learned, and the enhanced samples with the same probability distribution are generated on a large scale to solve the category imbalance problem. Finally, a gradient-weighted class activation mapping convolutional neural network is proposed to construct a transient stability assessment model to improve the reliability and interpretability of the model. The simulation results of IEEE 39-bus system test show that compared with other methods, the proposed model has higher stability discrimination accuracy, and the recognition rate of unstable samples is significantly improved, which verify the effectiveness of the proposed method.