Study On Corrosion Prediction Of Q235 Steel From Soil In Hainan Substation Based On Artificial Neural Network

Grounding grid is an important guarantee to ensure the safety of power system,power equipment and staff in substations. Because the grounding grid which is often made of Q235 steel, is buried in the earth all year round, its corrosion would degrade its grounding performance affecting the normal operation of the whole power grid.Therefore, it is important to study the corrosion status of grounding grid in the condition of non excavation and continuous electric power, and to find out the fault in time, which is of great significance to the safe operation of the substation.This paper is base on the domestic and foreign research on the corrosion of Q235 steel grounding grid and the project of Hainan Electric Power Technology Research Institute which is aimed at the soil corrosion in substation in an island environment especial for Hainan, of which research objects and contributions are met for the security requirement in the scene of substation in Hainan.Firstly, the corrosion theory for Q235 steel buried in substation soil is analyzed especially for the electrochemical corrosion forms and corrosion products. The relevant data of the physical and chemical properties are obtained by analysis the soil sample in labor and collecting in the filed, which provides a data foundation for the test of the later neural network model.Secondly, the basic principles of BP neural network and improved algorithm are introduced, at the same time, 2000 training samples and 200 test samples are randomly generated by unifrnd function among soil corrosion grade evaluation index, which solves the problem of small sample number and enhances the robustness and accuracy of sample identification. To select the best structural parameter of BP and RBF neural network for the more optimization and stability, method of formula and empirical test are used before the predicted results are contrasted of the different neural network model.At last, the data from the soil test of substation in Hainan Province are examined for the finished neural network above, of which predicted results are discussed. Comparative analysis showed that the the results from two neural network conform to the actual corrosion measurement. But the RBF neural network performs better in the constructed architecture, optimization and the whole accuracy.