Study On Mode Identification Of Low Frequency Oscillation In Power System Based On Intelligent Computing

With the rapid development of domestic economy,the power grid is expanding corresponding and interconnection of it is closely increased.Although interconnection of large area power grid may improve the economic benefit of system,it makes the dynamic system more complicated.Low frequency oscillation(LFO)limits power exchange and energy exchange between the regional power grid,even leads to accidents like system split or large area blackout,which is an important factor concerned the stability of power system.Mode analysis is the basis of LFO,the common method at present mainly includes the linear and nonlinear analysis method based on the model and signal analysis method based on the measured data.With the popular of the wide area measurement system(WAMS),signal analysis method based on the measured data has been widely applied.The paper applied three intelligent computing algorithms,singular value divide(SVD),group search optimizer(GSO)and deep belief networks(DBN)in LFO mode identification.The Prony’s method is a commonly used method in WAMS,this paper introduces in detail the principle and key points.On the basis of LFO model,the paper proposed a method based on SVD and GSO.SVD is used for determining the order of the signal,which is two times of the mode numbers.The paper also designed a generalized morphological filter(GMF),which is used for preprocess of the signal before carry out SVD.This filter improved the performance of the SVD even in environment with heavy noise.As for the parameter identification of LFO,the paper convert it to a minimum optimization problem with GSO.Due to strong global convergence of GSO,it is especially suitable for high dimension of multimodal optimization problems.Experiment shows that method in this paper has stronger noise resistance than traditional Prony algorithm,meanwhile,it has better convergence stability than PSO and GA algorithm.LFO is mainly divided into two types,regional and local oscillations.Two oscillations show difference from each other.The paper attempted to use DBN to identify the most vital dominate modal parameters in two kinds of oscillation scene.Experiment indicated a high realtime performance,a acceptable accuracy and a better resistance to noise.