<p>The roughness coefficient is a critical parameter in urban drainage pipeline design, and accurate estimation of this coefficient is crucial. Traditional empirical formulas have limitations in predicting the roughness coefficient, particularly for complex nonlinear problems. To overcome these limitations, this study conducted physical experiments to collect extensive hydraulic data. The pipe slope, bed roughness, hydraulic radius, and Reynolds number were used as input variables, and the roughness coefficient as the output variable, to construct a backpropagation neural network (BPNN) model. A genetic algorithm (GA) was used to optimize the weights and biases of the BPNN, resulting in a GA-BPNN model with enhanced prediction accuracy. The results show that the GA-BPNN model outperforms the traditional BPNN by exhibiting smaller error margins and better prediction accuracy across various conditions. Sensitivity analysis reveals that the Reynolds number and hydraulic radius exert a pronounced effect on the model’s predictions. This study offers new insights and methods for predicting the roughness coefficient of sediment-laden drainage pipes and for optimizing drainage pipeline design.</p>