With the continuous expansion of urban rail transit networks, equipment operation and maintenance work is facing increasingly severe challenges, mainly reflected in three aspects: heavy operation and maintenance load, limited efficiency improvement, and slow response speed. In the past, this field mainly relied on manual inspection methods, and technicians often relied on experience to set thresholds for fault diagnosis. The prediction accuracy of such methods was generally lower than 70%, and the average time from discovering abnormalities to dispatching maintenance personnel exceeded 8 hours. Therefore, the operation and maintenance costs have remained high for a long time, accounting for more than 30% of the total annual operating expenses. To address the above issues, this study has developed a smart operation and maintenance system based on the "cloud edge end" collaborative architecture. This system integrates various algorithms such as time series analysis models combining long short-term memory networks with Transformers, graph neural networks, and deep reinforcement learning, achieving intelligent coverage of the entire process from multi-source data collection, fault prediction and diagnosis, to dynamic maintenance strategy generation. To verify the effectiveness of the system, experimental analysis was conducted on the operation data of traction inverters, overhead contact systems, and signal machines on a certain subway line from 2023 to 2024. The results show that the fault prediction accuracy of the system has increased to 91%, which is 26 percentage points higher than the conventional threshold based methods and 11 percentage points higher than using the long short-term memory network model alone. At the same time, the maintenance response time has been shortened by 40%, and the annual operation and maintenance costs have correspondingly decreased by 25%.
Liu et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: