Application Of Naive Bayes Intelligent Algorithm For On-line Monitoring And Assessment Of Subsynchronous Oscillation In Power Systems

The subsynchronous oscillation of power system is one of the stability problems of power system,and the stability of power system is always a challenging research topic.In recent years,due to the increase of extra-high voltage,long-distance transmission lines and new energy grid-connected capacity in the power system,the series compensation technology and high-power power electronics technology have been widely used,and the new problems of subsynchronous oscillation caused by the technology have become increasingly prominent.The causes,manifestations,monitoring and suppression techniques,risk and impact assessment of a new round of subsynchronous oscillation have attracted worldwide attention once again.In this paper,three problems of mode parameter identification,on-line monitoring and evaluation of subsynchronous oscillation are studied.In view of the problems of noise sensitivity and low identification accuracy of traditional methods for parameter identification of subsynchronous oscillations,this paper proposes to apply synchrosqueezed wavelet transform(SST)to parameter identification of subsynchronous oscillations.SST combines the advantages of empirical mode decomposition(EMD)and wavelet,overcomes the problem of EMD mode aliasing,so it has better mode identification ability,in addition,the method has a certain anti-noise performance.Firstly,the time-frequency spectrum of the acquired subsynchronous oscillation signal is obtained by SST forward transformation.Secondly,the attenuation characteristics of each mode component in the spectrum are analyzed by using the automatic recognition algorithm proposed in this paper to determine the modes that need to be reconstructed.Then,only the divergent mode components are reconstructed.Finally,the reconstructed mode parameters are identified by Hilbert transform(HT)and least square method.At present,there are some problems in the on-line monitoring of subsynchronous oscillation based on PMU: parameter identification generally does not identify the attenuation factor,and the determination of alarm threshold requires human experience,which makes it difficult to guarantee the rapidity and reliability of the alarm criterion.In order to solve this problem,SET and NB are combined to monitor the subsynchronous oscillation.Compared with SST method,SET method has lower requirements on the sample frequency of signals,faster computation and is more suitable for parameter identification of subsynchronous oscillation signals on PMU.Firstly,the existing historical subsynchronous oscillation data are identified by SET,and the frequency and attenuation factors obtained from the identification are given to NB learning togenerate NB classifier.Then,when there are new PMU uploaded oscillation data,the SET is used to identify the oscillation frequency and attenuation factor,and then these parameters are sent to NB classifier to determine whether the subsynchronous oscillation occurs,it can achieve accurate warning,so as to realize the on-line monitoring of the subsynchronous oscillation.The risk assessment of subsynchronous resonance is an important consideration before the operation of changing the grid structure,such as grid-connection,planning and installing lines or series compensation capacitors.The accurate location of the resonance source is also needed for the design of the subsynchronous resonance suppressor on the system side.In order to solve this problem,modal analysis(MA)and naive bayes(NB)method are combined to evaluate the subsynchronous resonance.The modal analysis method can find the source of the subsynchronous resonance by decoupling the system loop impedance matrix,and then obtain the information of the lines,components and propagation distance involved in the resonance.However,in the process of modal analysis,the subsynchronous resonance frequency and the participation factor in the loop need to be judged by human participation.In this paper,naive bayes method is adopted to replace the part of human participation.Two NB classifiers are obtained through the training of NB to judge the subsynchronous resonance frequency and the participation factor in the loop,so that the modal analysis method combined with NB has the ability of intelligent evaluation of subsynchronous resonance.