Research On Methods For Power System Transient Stability Assessment Based On Machine Learning

Power system transient stability assessment is critical to maintain the safe and stable operation of power system.With the expansion of modern power system scale and the increase of system operation uncertainty,transient stability assessment is facing greater challenges in terms of assessment accuracy and calculation speed.The rapid development of machine learning has injected new vitality into transient stability assessment,and transient stability assessment methods based on machine learning models have become a research hotspot in power system.The thesis has carried out research on this.The main work and achievements are summarized as follows:Some key problems in the basic process of establishing machine learning model for transient stability assessment,including data processing,feature selection and model evaluation index are studied.The necessity of data transformation for transient stability assessment samples is discussed.The sampling method is proposed to improve the category imbalance problem in transient stability assessment samples.The influence of different input characteristics on the model is analyzed.The model evaluation index suitable for unbalanced classification problems is studied.The above research is verified and lays the foundation for the following research content.A power system transient stability assessment method based on clustering adaptive active learning is proposed.The method uses K-means algorithm to select the initial samples of active learning,making the initial samples representative and speeding up the active learning process.In order to select the sample which improves the model the most in each iteration,a parameter adaptive method combining uncertainty index and representative index is proposed.The sample uncertainty index is measured by the size of the entropy value,and the sample representative index is measured by the Euclidean distance between samples.In the process of active learning,the weights of the two indexes are adjusted dynamically according to the change of the accuracy of the model after adding the new sample to select the best sample.The simulation results show that the proposed method can significantly reduce the number of labeled samples and the time required for time-domain simulation under the premise of ensuring the accuracy of the model,thus reducing the total time-consuming of sample preparation and model training.Compared with the traditional active learning selection strategy,the proposed method greatly improves training efficiency.A power system transient stability assessment method based on graph convolutional network is proposed.The topological information of the power system is taken as part of the input,and the node information and topological information are aggregated.A jumping knowledge layer is added to improve the over smoothing problem caused by too many graph convolutional layers.The simulation results show that the performance of the proposed method is better than that of the machine learning model without considering the power system topology information.It has strong adaptability to the system with topology changes,and the old model can be fine-tuned to adapt to the topology of the new system quickly with good generalization ability.