Small Signal Stability Assessment And Early Warning Of Power System Based On Data Driven Methods

Small signal stability is the key factor of power system's safety operation.However,the interconnection of power system and the connection of new energy bring the challenge of fast variety and fluctuation to power system,which requires the assessment and early warning of small signal stability accomplished in shorter time.Traditional method of computing key eigenvalues is complex and time-consuming due to the increasing dimension of dynamic model of power system.Therefore,a fast method of small signal stability assessment and early warning is required.Meanwhile,with the development of smart grid,a great deal of system operation data has been produced.Data driven methods can not only utilize the historical data,but also ensure the speed of small signal stability assessment and early warning.This paper firstly proposes a small-signal stability assessment method of power system based on convolutional neural network.This method takes wide-area monitoring system(WAMS)signals as model input and generates critical eigenvalues as output.After the necessary preprocessing of the input and output,the mapping relationship between input and output can be established by deep neural network.Discrete cosine transformation(DCT)and graphics processing unit(GPU)parallelization techniques are employed,in order to overcome the challenges from high system dimension and slow training rate.Case study results indicate that the proposed method is able to accurately obtain critical eigenvalues of the studied system after offline training using historic data,given no significant change in control parameters.Furthermore,this method can deal with topological change and data loss.Then,based on the CNN,this paper introduces the changing information of power system in time line,and proposes a small-signal stability early warning method of power system based on two-stream convolutional neural network.This method takes the power signals detected by wide-area monitoring system(WAMS)and the changing information of power signals as model input and generates the movement vector of critical eigenvalues as output.After the necessary preprocessing of the input and output,the mapping relationship between input and output can be established by the two-stream convolutional neural network.In order to overcome the challenges from the high time and space complexity of two-stream convolutional neural network,a new structure of two-stream convolutional neural network is proposed combining with CNN model compression techniques,which is suitable for small signal stability early warning.Case study results indicate that the proposed method can accurately forecast the movement vector of key eigenvalues after offline training using historic data,given the loads fluctuate regularly.