Data Augment Method for Power System Transient Stability Assessment Based on DeepSMOTE

Data-driven transient stability assessment (TSA) method has become the focus of power system stability and security research in recent years. However, in actual power system, transient unstable data only accounts for a small portion of all data. Unbalanced data is not conducive to the machine learning model's judgment of instability situations. A data augment method based on DeepSMOTE is proposed to generate high-quality unstable datasets. Then, an extreme learning machine (ELM) classifier is used to select the generated dataset with the highest G-mean value, which is used to fuse with the original dataset to construct a balanced dataset with a similar number of transient stability and instability samples. Finally, TSA is implemented based on the balanced dataset. It has been verified in the modified IEEE 39-bus system simulation example containing new energy that DeepSMOTE can effectively generate data that follows the distribution characteristics of the original data of power system, thereby improving the recognition rate of unstable samples in TSA. Moreover, it has strong robustness under wide area noise and effectively balances power system instability data.