Risk Detection And Source Tracing Of Subsynchronous Oscillation In A Power System Integrate With Wind Power Generation Based On Transfer Learning

Oscillation instability caused by grid connection of wind power is threatens the safe operation of modern power systems,and oscillation risk detection and source tracing are the basis for the implementation of oscillation suppression strategies.At present,the oscillation risk detection and source tracing methods of wind power grid-connected systems faces three major challenges:First,the large number of wind turbine generators in large-scale wind power aggregation systems and the difficulty of obtaining detailed parameters make it difficult for conventional stability analysis methods based on parametric models to be useful.Secondly,in the practical system,the measured data presents a data imbalance state with sufficient steady-state data and lack of oscillation data,and the data samples lack oscillation labels,which are difficult to be directly apply to data-driven modeling.Thirdly,although the simulation system model can be established,the generalization of the data-driven model trained by the simulation system is insufficient due to the differences between the simulation system and the practical system.In order to solve the above problems,this paper studies the risk detection and source tracing method of subsynchronous oscillation of wind power grid-connected system based on transfer learning,and the main work and innovative achievements include:(1)The induced mechanism of subsynchronous oscillation of wind power gridconnected system is analyzed,and the mapping relationship between measured data and oscillation mode or participation factor is established.On the basis of demonstrating the feasibility of feature-based transfer learning,the application paradigm of transfer learning in the oscillation risk identification of wind power grid-connected system is proposed for the first time.Under this paradigm,the oscillation risk detection and source tracing methods are studied for different subsynchronous oscillation scenarios of wind power grid-connected systems.(2)Aiming at the problem of subsynchronous oscillation risk detection in largescale wind power grid-connected system,a data-driven method for oscillation stability assessment based on dynamic parameter equivalence is proposed.This method can transform the modeling of large-scale wind farm under the condition of unknown wind turbine parameters into the time domain simulation of equivalent simulation system,so as to transfer the calculation burden of online stability assessment to the offline datadriven modeling stage.and a data-driven model for oscillatory stability assessment based on transfer learning is established.The example test shows that the oscillation stability assessment of wind power grid-connected system can be realized by the measured data in the original large-scale system,which avoids large-scale parametric modeling and high-dimensional matrix operation.(3)Aiming at the problem of subsynchronous oscillation source tracing in largescale wind power grid-connected system,an oscillation source inversion method in equivalent system and a batch generation method of training samples for deep learning model are proposed based on open-loop mode resonance theory.Taking two typical scenarios of subsynchronous oscillation caused by dynamic interaction between wind farm and synchronous machine,and dynamic interaction between wind farm and VSCHVDC as examples,the oscillation source tracing model based on joint adaptive transfer network and the oscillation source tracing model based on adversarial transfer network are developed respectively,and the performance of the tracing model is tested by different test sets.The research shows that the deep transfer network can minimize the feature difference between the simulation system and the practical system,so that the tracing model trained in the equivalent small-scale simulation system can still maintain the efficiency and accuracy of the model in the practical system.(4)In the large-scale wind farm grid-connected system,the wind farm may interact with different dynamic components under different operating conditions.To further expand the subsynchronous oscillation source tracing method of single scene in(3),a subsynchronous oscillation source tracing model based on multi-source domain adaptive network is developed,which can realize the subsynchronous oscillation source tracing in different evoked scenarios of complex power system integrate with wind power generation.The research shows that this model has stronger adaptability and stronger generalization for different induced scenes compared with other deep learning models,which lays a foundation for engineering applications for complex scenarios.