Research On Estimation Method Of Transient Overvoltage Of HVDC Sending-end Based On Control System Dynamic Process And Convolutional Neural Network

In recent years,with the expansion of the interconnection scale of power systems and the increase of DC transmission capacity,the stability problems of power systems have become increasingly prominent,especially when large DC power disturbances occur(such as DC block,commutation failure,etc.).The occurrence of transient overvoltages will cause a large impact on the transmission equipment on the AC and DC sides,which will seriously affect the stability of the power system.In view of the above problems,the transient overvoltage estimation method based on dynamic process of DC control system and the transient overvoltage estimation method based on convolutional neural network(CNN)are respectively considered in this paper.The main research contents are as follows:(1)Aiming at the lack of related research on the dynamic interaction of component control systems when traditional methods are used to estimate transient overvoltages,which leads to inaccurate estimation results,a transient overvoltage estimation taking into account the dynamic process of the control system is proposed method.First,analyze the dynamic characteristics of the system components at the sending end under the disturbance of AC and DC faults,obtain the main links and mechanisms that affect transient overvoltage,and establish an effective mapping of transient overvoltages and control links.Then,by comparing and analyzing the influence of different control parameters on transient overvoltage,determine the key factors affecting transient overvoltage.Finally,by deriving the transfer function,the mathematical model of the dynamic process of the control system of each power grid component during AC and DC faults is detailed to reduce the error of transient overvoltage estimation.(2)In view of the need to simplify the complex mathematical model of the power system when constructing the dynamic process of the grid element control system to improve the transient overvoltage estimation accuracy,it is difficult to construct the transient overvoltage estimation model.Therefore,this paper proposes a method estimating transient overvoltage based on CNN,which can avoid the detailed model construction process and has high evaluation accuracy.Firstly,based on the basic principles of feature construction according to CNN,a nonlinear network structure with multiple hidden layers is set up.By splicing the voltage,phase angle and power of each node collected by wide area measurement devices(WAMS)according to the topology relationship of grid nodes and the time sequence from fault occurrence to cutting-off,a matrix representing power grid state is obtained.Secondly,the hyper-parameters of CNN are optimized using gradient descent method for supervised training.Automatic extraction of key eigenvalues is realized by optimizing the weight matrix between input layer and convolution layer.Finally,using CNN's deep architecture,a mapping model between transient overvoltage and input data is constructed,and the transient overvoltage of DC-transmitted system is estimated promptly and accurately.