Automatic Calibration Framework of SWMM Parameters Based on SSA-Optimized BP Neural Network

Abstract

The storm water management model (SWMM) incorporates numerous parameters that are difficult to measure directly and whose calibration process is complex and time-consuming, significantly limiting the accuracy and efficiency of urban stormwater process simulation. In response to this problem, this study proposes a new parameter calibration framework for the SWMM model based on the sparrow search algorithm (SSA) and the backpropagation neural network (BPNN). The framework utilizes SSA to provide the optimal initial weights and thresholds for BPNN, constructing an efficient and stable SSA-BPNN surrogate model. This surrogate model can reflect the relationship between the uncertain parameters of the SWMM model and the water depth at nodes, thereby enabling the rapid selection of an appropriate parameter combination for the SWMM model and the improvement of the model’s fitting accuracy. Taking a certain urban subbasin in the Zhuzhou urban area of Hunan Province as the research object, seven rainfall events with different characteristics were selected for model calibration and verification. The results indicate that the probability of the SSA-BPNN obtaining effective calibration results with all parameters within a reasonable range is 89.6%, which is significantly higher than that of BPNN at 42.8%. In terms of calibration accuracy, the average values of the coefficient of Nash–Sutcliffe efficiency, the determination coefficient (R2), and the relative error of peak node water depth (REp) for the SSA-BPNN are 0.916%, 0.958%, and 3.164%, respectively, outperforming manual calibration, BPNN, and particle swarm optimization combined BPNN. Additionally, the SSA-BPNN can maintain good performance under different rainfall intensities, durations, and multipeak rainfall events, significantly enhancing the model’s simulation ability and providing a reference for urban storm-flood simulation and sustainable urban water management in practice.