Analysis Of Low Frequency Oscillations Based On Deep Learning

With the rapid development of economy and power gird,the structure of power gird is more complex.Low frequency oscillation of large-scale interconnected power systems may result in equipment damage,which threats the safe operation of the power grid seriously.Identification of oscillation pattern is important for suppression of low frequency oscillation.Prony algorithm has the advantage of convenient,high accuracy and fast calculation,which is the common method for extracting feature of low frequency oscillation.According to the integration of deep learning algorithm and Prony algorithm,this paper proposed a novel algorithm for analysis of low frequency oscillation based on deep learning,which solves the problem of order selection and noise immunity.Above all,this paper introduces the basic principle of Prony algorithm and mathematical derivation of the model,and studies the influence of fitting precision of Prony algorithm on order selection and noise.The simulation reveal that lower model order selection causes a lower precision in identification of low frequency oscillation;the higher order selection increases computational cost and the number of disturbance mode.A FIR lowpass filter method is presented for improve Prony algorithmic poor noise immunity.The simulation results show that this method is effective.Afterwards,this paper introduces the theory of deep learning and proposed a novel algorithm for analysis of low frequency oscillation's order based on deep learning.The algorithm is consist of two parts: deep belief network(DBN)which is consist of three restricted Boltzamann machines(RBM),and softmax classifier which classify order of low frequency oscillation.The proposed algorithm include three steps: firstly,training the model based on training samples;secondly,obtaining probability distribution of the identifying low frequency oscillation's order;thirdly,obtaining order of low frequency oscillation based on the probability distribution.Simulation results indicate that the proposed algorithm identifies order of low frequency oscillation exactly,which has advantage of identifying high order of low frequency oscillation,compared with the traditional SVD method and BP neural networks.Finally,this paper present a novel method for analysis of low frequency oscillation based on deep learning.The main steps of the proposed method are as follows.The deep learning model is trained based on training samples firstly,and then the order of the identifying low frequency oscillation preprocessed by FIR lowpass filter is obtained based on deep learning.Finally low frequency oscillation is estimated by Prony algorithm based on the order.Theproposed method is applied in 10-generator,39-bus system simulated by RTDS to analyze low frequency oscillation.The MATLAB simulation showed the effective of the proposed method.