Engineering Reports General Engineering

Research on a Three‐Level Government Emergency Resource Dynamic Scheduling Model Based on an Improved PPO Algorithm Under Extreme Blizzard Conditions

Yixuan Mu, Miao Zhang
Wiley · 2026-11-01 · DOI 10.1002/eng2.71107

Abstract

ABSTRACT Extreme blizzards simultaneously disrupt road accessibility, prolong transportation time, and amplify heterogeneous demand, creating a dynamic coordination problem for emergency resources transferred through central reserve, provincial transit, and municipal distribution levels. This study develops a task‐oriented Hierarchical‐Attention Proximal Policy Optimization method (LA‐PPO) for aggregate three‐level resource scheduling. The state representation integrates multi‐level inventories, road icing and blockage conditions, city‐level unmet demand, and in‐transit resources. A lightweight hierarchical attention module adaptively fuses central‐, provincial‐, and municipal‐level information, while a state‐dependent reward coordinates timeliness, fairness, economy, and robustness through a researcher‐specified indicator‐to‐weight mapping. Under the experimental scope defined in this study, LA‐PPO reduced total response time from 5.1 h to 4.2 h and average delivery delay from 22.6 min to 15.3 min relative to traditional PPO in the basic static scenario, with a Gini coefficient of 0.18 and resource utilization of 90%. At a 20% road‐blockage level, it achieved a 95% scheduling success rate and a 3.5 min adjustment time. In the 2023 Northeast China historical replay, total response time was 4.1 h, resource utilization was 92%, and demand satisfaction was 95%. Ablation results further showed 4.2 h ± 0.1 h response time and 95% ± 1% blockage‐scenario success for LA‐PPO, compared with 5.0 h ± 0.3 h and 83% ± 2% for direct state concatenation. These results support the value of hierarchical state fusion for the specified scheduling task, but they do not constitute validation of a complete governmental emergency‐management process or live operational deployment.

Read at publisher →