Flood Forecasting Scheme Based on Artificial Neural Networks

Abstract

Flood prediction is of great significance for safeguarding lives and property. Existing methods mostly employ linear models such as logistic regression, which, despite their simplicity, struggle to capture the nonlinear relationships between multiple factors and flood risk, resulting in limited predictive accuracy. This paper constructs a flood prediction model based on a BP neural network, selecting nine flood impact factors and normalizing them, with logistic regression as a benchmark model. The model performance is comprehensively evaluated on the test set in terms of R 2 , RMSE, and accuracy. Results show that the BP neural network achieves an R 2 of 0.9958 and an RMSE of only 0.0163 on the test set, far superior to the logistic regression model (R 2 =0.7187, RMSE = 0.1338), verifying the effectiveness of BP neural networks in flood prediction.