Research On Passive Islanding Detection Method For Synchronous Distributed Generation

With the rapid depletion of traditional fossil fuels and its serious impact on environmental pollution,people hope to replace traditional fossil fuels by clean energy such as wind energy and solar energy.Distributed generation technology is to use various clean energy sources on load side to generate electricity.It has become a hot research topic in recent years because of its high energy utilization,high reliability of power supply and environmental protection.Unintended islanding detection of distributed generation is the basis to ensure its safe and stable operation.For this reason,the islanding detection of synchronous distributed generation is studied in this paper.The main work and achievements are summarized as follows:Firstly,the mathematical model of islanding event of synchronous distributed power generation is established.The characteristics of frequency,frequency change rate and rotor angle change with time after islanding event of synchronous distributed power generation are obtained,and the influencing factors are analyzed.The characteristics of over/under-frequency islanding protection are verified by analysis and simulation.The concept of non-detection zone of synchronous distributed power generation is given.The non-detection zone of distributed power generation is obtained in typical distributed power system.Secondly,a method of islanding detection for distributed power supply based on LSTM is proposed.Considering the time-varying characteristics of frequency and voltage of synchronous distributed generators after islanding events,a deep learning model is proposed,which can effectively process time series signals.The structure of LSTM model is optimized,three typical LSTM model structures are obtained,and their performances are analyzed.The simulation results show that too many model parameters will lead to over-fitting,and too few model parameters will lead to under-fitting.The detection accuracy of the proposed method is compared with that of the existing distributed power supply islanding detection method,and compared with the traditional over/under frequency islanding protection.The simulation results show that the proposed method can effectively improve the accuracy of islanding detection of distributed power supply and reduce its non-detection zone.Finally,a method of islanding detection based on transfer learning is proposed.This method can transfer the existing LSTM model of islanding detection of distributed power supply to other distributed power supply when the training data set is too small,so as to meet the requirements of practical application.The parameters of the LSTM layer that should be retained and the LSTM layer that should be transferred are analyzed and verified.The results show that the first two parameters of the existing model should be retained in the transfer process.In addition,the performance of the model directly trained with a small number of training data sets and the model based on migration learning are compared and analyzed.The simulation results show that the performance of the model based on transfer learning is better than that of the model directly trained with a small number of training data sets.