Analysis of Runoff Coefficient Prediction Based on LM-BP Neural Network

Abstract

Abstract Runoff coefficients are affected by many factors, and their complex nonlinear relationships make it difficult to calculate accurate runoff coefficients using experimental physical models. In this paper, we improved the traditional BP neural network model based on the Levenberg-Marquardt method and established an S-type/S-type mathematical model of the relationship between runoff coefficients and influencing factors to predict each surface runoff coefficient under different rainfall conditions and different subsurface conditions, and compared it with other methods. The results calculated after the actual case simulation showed that the error of LM-BP neural network simulation was within the range of 0.03~0.09, the error was smaller, the calculation results were more accurate, and the prediction of runoff coefficient had the advantages of strong generalization ability and high prediction accuracy, which was a great improvement to the traditional rainfall-runoff coefficient best-fit relational fitting relationship method. Besides, in order to reduce the problem of sample overtraining, the more detailed the hydrological information, the better.