| The power system is developing in the direction of high proportion of renewable energy,high proportion of power electronic equipment,multi-energy complementary comprehensive energy utilization and deep integration of physical information.The development of low-inertia renewable energy and the complex interconnection of large power grids lead to the modern power system exhibiting strong time-variability,strong nonlinearity,strong uncertainty and data diversity.The dynamic behavior of system frequency after disturbance is more and more complex.In power system frequency operation and control,it is necessary to comprehensively analyze the frequency security and stability state,the frequency dynamic behavior characteristics,and the disturbance response information recorded at each measurement point to assess the system frequency security and stability,and to provide auxiliary for the comprehensive decision of frequency operation and scheduling as well as the selection of control methods.It is a challenging research task to obtain fast and accurate information about the macroscopic properties and local correlation of the system based on the observed system state.There are many limitations to simply describe the problem characteristics of the power system frequency under the new form by using a mathematical model.As an emerging Artificial Intelligence(AI)technology path,deep learning(DL)has unique advantages in complex problems such as power system frequency analysis and control with its powerful data analysis,prediction,and classification capabilities.Therefore,it is important to study the theory of AI-assisted frequency dynamic analysis of power systems in order to meet the challenges brought about by changes in power grid morphology,improve the adaptive capability of frequency response analysis and stability control,enrich the theoretical system and methodological means of frequency dynamic analysis,and guarantee system frequency stability.In this paper,the starting point of this paper is to improve the understanding of the frequency response characteristics of large-scale renewable energy and continuous UHVDC connection.The quantitative description of the frequency response pattren,the coupling mechanism of frequency characteristics and power angle oscillation,and intelligent methods such as deep learning and frequency response characteristic evaluation problem is combined to carry out research,and the main work is as follows.(1)For the problem of quantitative analysis of frequency response,a definition of frequency response pattern and its quantitative description and calculation method are proposed.Firstly,the dialectical relationship between dispersion,uniformity,independence and coupling of frequency response is systematically analyzed.Secondly,based on the characteristics of frequency dynamic behavior after perturbation,the characteristic quantity of frequency response is calculated and extracted from three aspects:dispersion index,uniformity index and systematic index;the calculation methods of dispersion index matrix and uniformity index matrix are proposed,the frequency response pattern is defined and constructed,and its quantitative description method is proposed.Finally,based on the proposed frequency response pattern,the impact of renewable energy penetration rate and UHVDC transmission on the frequency response pattern dispersion and uniformity indexes is analyzed through large grid case simulation.The quantitative information of frequency response pattern can establish the theoretical basis and quantitative analysis for the application of DL and other AI methods to frequency problems,and provide systematic theoretical and methodological support for the selection of intelligent methods to extract feature quantities.(2)Aiming at the phenomenon that the coupling between frequency response and other physical quantities is affected by its dispersion,a method is proposed to analyze the coupling characteristics and mechanism of frequency spatio-temporal distribution response and power angle oscillation.Firstly,a two-machine system model is established,the variation law of the system power-angle and the two-machine frequency after the disturbance is deduced,and the coupling mechanism of the frequency response characteristic and the power-angle characteristic under the two conditions of stability and instability of the system after the disturbance is analyzed.Then,an analysis framework of frequency-angle dynamic response coupling characteristics is constructed,and a quantitative index of power-angle oscillation characteristics is proposed to quantitatively describe the coupling characteristics of frequency response pattern and power-angle oscillation,and a quantitative evaluation index of coupling strength is proposed.Finally,the change process of the frequency response pattern during the power-angle oscillation is analyzed through a numerical example,the influence of the power-angle oscillation characteristics on the frequency response pattern is revealed,and the coupling strength of different frequency response indicators and power-angle indicators is quantitatively evaluated.Theoretical analysis and simulation show that the characteristic index makes the coupling characteristics and coupling strength of the power-angle oscillation and frequency response pattern quantifiable,which provides a reference for the subsequent in-depth study of the spatial-temporal distribution characteristics of the frequency response and optimal control measures.(3)Aiming at the problem of combining the spatial-temporal distribution characteristics of frequency response with DL,an artificial intelligence-assisted frequency response pattern evaluation method is proposed to achieve rapid and accurate evaluation of system frequency response pattern characteristics after large disturbances.Firstly,the adaptability of Convolutional Long Short-Term Memory(ConvLSTM)to frequency problems is discussed in depth,and a deep model considering the spatiotemporal distribution characteristics of frequency is constructed.Then,based on the frequency response analysis theory and the quantitative description of the frequency response pattern,the key feature variables are selected as the input and output features of the proposed model,and the selection of input feature categories is realized by intelligent methods.Realize a three-dimensional feature tensor construction method suitable for ConvLSTM considering spatial-temporal distribution characteristics.A PSS/E-based batch generation method for the frequency response pattern database of power system after disturbance is proposed.Finally,the network structure configuration and parameter tuning method of the proposed intelligent evaluation model are systematically studied to establish the deep network architecture and parameter configuration applicable to the frequency response pattern evaluation problem and improve the evaluation performance,generalization capability and robustness of the model.The analysis results of the actual large grid shows that the proposed model can evaluate the frequency response characteristic indexes more quickly and accurately,and has superior functions,better evaluation performance,stronger generalization ability and robustness compared with other DL models. |
PDF Full Text Request