Rule Extraction-based Data Augmentation Method for Transient Instability Identification of Power Systems Using Machine Learning

Machine learning methods have become a promising method for rapid identification of transient stability and unstable generators of power systems. However, the transient unstable instances are always fewer than stable instances, which brings difficulties of transient instability identification. This paper presents a data augmentation approach to increase the number of unstable instances. Firstly, original data will be generated by time domain simulation considering the randomness and limitations of power systems. Then, the patient rule induction space (Prism) method is used to obtain the unstable rule set from the training data. After that, new instances can be generated by the unstable rules and time domain simulation. Then, the machine learning method is used to generate the transient stability prediction model and unstable generator identification model based these two kinds of instances. All the generated instances can be verified by time domain simulation. Comprehensive studies are conducted on the New England 10-machine 39-bus system to verify the effectiveness of the proposed approach.