Research On Power System Dynamic Stability Evaluation Strategy Based On Deep Learning

The dynamic stability of the power system is an important indicator of the safe and stable operation of the power grid.However,the regional interconnection in the modern power system and a large number of new energy grid connections have brought new risks and challenges to the dynamic stability of the power grid.On the one hand,the regional interconnection of the power system through UHV AC/DC greatly increases the complexity of the system.On the other hand,a large number of new energy units are connected to the grid,which enhances the random fluctuation of the system and weakens the inertia and frequency control ability of the system.In recent years,the upsurge of deep learning has brought new ideas to the analysis of power system security and stability.The deep learning algorithm represented by convolution neural network has been successfully applied in processing image data with spatial characteristics.This thesis focuses on the prediction of power system oscillation frequency and damping ratio,and carries out the research of dynamic characteristic oscillation frequency and characteristic damping ratio prediction of power system after disturbance based on CNN network;Aiming at the advantages of physical model and machine learning method,the prediction method of oscillation frequency and damping ratio based on CNN-LSTM is studied.The main research contents of the thesis are summarized as follows:Firstly,this thesis studies the basic concepts and theories of power system dynamic stability.Based on various low-frequency oscillation research theories,a single machine infinite bus model for power system dynamic stability is established,and the dynamic stability of the power system is analyzed and solved by solving nonlinear equations.The analysis method for constructing a linearized model and obtaining eigenvectors is determined.The traditional Prony prediction method for dynamic stability was studied from the aspects of model order determination,noise impact,and ill conditioned equation problems,and several limitations of the Prony method were analyzed.Secondly,this thesis proposes a method for screening the input characteristics of CNN network and generating the system operation state database.Based on the process of power system oscillation frequency and damping ratio under power disturbance,the key power system operation data information is screened as the input characteristic of the CNN,and the oscillation frequency and damping ratio of the system after disturbance are used as the output.The training of frequency analysis model based on CNN is completed by using tensor sample data.Taking an actual power grid in China as an example,the anti-noise and accuracy of this method are proved by comparing the traditional Prony analysis method with other deep learning methods.Finally,aiming at the advantages and disadvantages of physical model and deep learning model for power system oscillation frequency and damping ratio prediction,a physical-information fusion method for power system oscillation frequency and damping ratio prediction after disturbance is proposed.First of all,an equivalent physical model of the power system taking into account the doubly-fed fans is used to establish the key physical relationship between the power system disturbance and the oscillation frequency and damping ratio,and then the CNN-LSTM network is used to establish a deep learning model based on the power system operating state information.Finally,the adaptive neuro-fuzzy inference system is used to organically fuse the frequency prediction results of the two sub-models to achieve integrated learning,The lowest frequency prediction result after physical-information fusion is obtained.Finally,the accuracy of the proposed method is verified by the actual system of a power grid in China,and the impact of wind power grid connection on the dynamic stability of the power grid is analyzed.