Analyzing drainage system state through pipeline network complexity and driver interactions using multiscale entropy and deep learning

Quantification of the state of the drainage system is essential for the prevention and management of urban flooding. Conventional approaches typically employ various indicators with subjective weighting. Yet, due to the inherent complexity of urban drainage systems, these methods often fall short in objectively and comprehensively encapsulating the inherent complexity of drainage systems. Pipelines are integral to drainage systems, yet existing research has predominantly examined their physical attributes, overlooking the connection between pipeline dynamics, ground conditions, and overall system processes. This study presents an innovative method that leverages pipeline network complexity to depict the drainage system’s status, this approach serves to bridge the gap between pipeline network dynamics and the functionality of the drainage system. The Pipeline Network Refined Complexity Multiple Dispersion Entropy (PN_RCMDE) method was developed to quantify pipeline complexity across various time scales. The study utilizes the Pipeline Network Complexity Informer Model (PNCInformer) in conjunction with Shapley Additive Explanations (SHAP) to investigate the interplay between pipeline network complexity and influencing variables. This approach is implemented in a specific region within Anhui Province, China, which is susceptible to flooding. The results revealed that the network complexity fluctuates significantly during peak rainfall periods, and that the main influencing factors differ from region to region, thus requiring tailor-made management strategies. Nodes should be optimized and storage facilities should be increased in the Outfall_A linkage area. Outfall_B interconnection areas need to optimize the pipeline, while taking care to prevent node overflows. Outfall_C interconnection area needs to use pipeline expansion to reduce the impact of hydraulic loads on the network. Outfall_D interconnection area should be functionally upgraded and structurally optimized at the nodes to reduce the negative impacts of peak rainfall. • A novel method uses pipelines network complexity to assess drainage system state. • The PN_RCMDE method was developed to capture multi-scale pipelines network dynamic. • The PNCInformer model linked pipelines network to its influencing factors. • Suggested management strategies based on spatial–temporal driver analysis.