Establishing Neural Network Models for Predicting Flood Propagation and Recession in Urban Roads

Abstract

In this analysis, a contagion model is a straightforward yet effective mathematical approach—that was used to forecast the temporal change of the outset and contiguous distribution and recession of flooding in metropolitan roadway networks. A system of metropolitan roadways must be flood-resistant in order to provide public services and deal with emergencies. The dispersion of floodwaters is a complicated temporal-spatial process that affects urban networks. In comparison to the SEIR (Susceptible-Exposed-Infected-Recovered) prototype, a system of ordinary differential equations, four macroscopic characteristics, rate of including flood spreading represented by ($\beta$), rate of flood incubation symbolized by ($\alpha$), and rate of recovery highlighted by ($\mu$), can be used to understand how floods evolve within networks. Additionally, by joining the backpropagated neural network and the Levenberg—Marquardt algorithm (NN-BLMA), surrogate solutions to the model are discovered. This method has some clear advantages over conventional ones, including flexibility, comparatively simple implementation, and fastest results. Reference solutions are generated using the Runge-Kutta of order four (RK4) method. We have examined three distinct scenarios to analyze our surrogate solution models. By changing $\alpha$, $\beta$, $\mu$, and k, the mathematical model's stability and equilibrium are examined. To gauge the validity of our machine learning process, we categorize our candidate solutions into training, testing, and experimental class. The efficacy of the NN-BLMA scheme has been confirmed by comparative examinations of statistical values based on mean squared error function (MSEF), effectiveness, regression plots, and failure histograms.