Power System Transient Stability Assessment Based on Bayesian Optimized LightGBM

Abstract

In case of dealing with large-scale and high dimension data samples, one of the key challenges for the artificial intelligence (AI) based transient stability assessment, is to guarantee the offline training speed and quickly determine the optimal parameters of the adopted algorithm. To cope with this issue, a transient stability assessment method based on Bayesian optimized Light GBM is proposed in this paper. This approach uses gradient-based one side sampling (GOSS), histogram algorithm and leaf-wise with depth restrictions to accelerate the model training process, during that the optimal parameters are quickly determined leveraging the Bayesian optimization. To ease the model complexity, the importance of the input data’s features is explicitly determined during the process of generating the decision tree. Combined with the correlation analysis, the relationship between the input data and the transient stability of considered contingency sets can be excavated more intuitively to select the necessary features. Test results on the New England 39-bus system show that the proposed approach has higher accuracy and speed, as well as a higher recognition rate for unstable samples compared with other machine learning based methods.