Grounding refers to the electrical connection between nodes or some conductive parts of electrical equipment in power communication and the earth. It is a channel for conducting current, which can ensure that the potential at the electrical connection is kept within a certain allowable range. As an important part of power generation and substation, grounding grid is the fundamental guarantee and important measure to maintain the safe and reliable operation of power system and ensure the safety of operators and electrical equipment. Its grounding performance has been widely concerned by power workers at home and abroad. Because the conductors constituting the grounding grid are buried underground, the conductors and grounding leads of the grounding grid are often corroded or even broken due to poor welding, missing welding, soil corrosion, grounding short circuit, current electrodynamics and other reasons, resulting in the deterioration of the electrical connection performance of the grounding grid. Therefore, it is of great significance to study the fault diagnosis of grounding grid. In this paper, the fault diagnosis of grounding grid in power plant and substation is studied. Based on the analysis of various fault characteristics and existing fault diagnosis methods, a variable model of grounding grid parameters considering inductance and capacitance is proposed. Combining wavelet transform with neural network, the strategy of fault feature extraction and fault identification of grounding grid and its implementation method are put forward, thus improving the accuracy and reliability of fault diagnosis of grounding grid. The paper mainly includes the following aspects: Based on EMTP electromagnetic simulation software, a transmission line parameter model composed of resistance, inductance and capacitance is established. The topological structure and parameters of the model can be configured according to the actual calculation needs, which provides effective simulation data support for the research of grounding grid fault diagnosis. The simulation results prove the feasibility of the model. The influence of excitation source and branch fault position on the change of node voltage waveform is studied, and it is proposed that the measurement error caused by branch direction can be avoided by using oblique excitation source in simulation model. It is proposed to extract fault features by using high-frequency excitation through wavelet transform, that is, first, high-frequency excitation is applied to the grounding grid to obtain the node voltage waveform, and the signals before and after the fault are decomposed and transformed by using db3 wavelet to construct the comprehensive energy feature vector of the grounding grid branch fault. On this basis, the fault diagnosis method of grounding network based on neural network is given, and the simulation results prove the effectiveness and accuracy of this method.