Restoration method of DC ground potential distribution law based on neural network

DC Surface Potential (DCESP) distribution is the key to DC bias in power system transformers. Unbalanced or unipolar operation of the HVDC transmission system, and the stray current of the rail transit traction power supply will affect the distribution of DCESP. The DC surface potential distribution is affected by the magnitude and location of the DC in-ground current, as well as the soil structure and parameters. In fact, whether it is a DC transmission system, a rail transit system, or an AC grid system, its geographical dimensions are measured in kilometers, ranging from a dozen kilometers to thousands of kilometers. How to measure the actual DC surface potential distribution is of great significance for studying DC bias. The terrain between the existing two substations is complex, especially in cities affected by roads, building and other factors. The number of locations available for measuring surface potential data is limited and discontinuous. In this paper, a method for restoring DC surface potential distribution based on data from discontinuous locations is studied. Firstly, a CDEGS simulation model was built, and the DC surface potential distribution data around the subway was obtained by simulation, and data from several discontinuous sections were intercepted as the simulated measurement sections. Using the relative potential of the measurement section as a known quantity to train a built neural network model, which is a neural network with a dynamic feedback structure. Finally, the remaining test data is substituted into the trained neural network for testing, and its reduction effect and influencing factors are analyzed. The research results in this paper provide a new idea for DC surface potential measurement and can promote the development of DC bias magnetic field.