📄 中文摘要
桥梁定期检查记录包含有关公共基础设施的敏感信息，使得在现有数据治理约束下跨组织的数据共享变得不切实际。提出了一种联邦框架，用于估计桥梁劣化的连续时间马尔可夫链（CTMC）风险模型，使各市能够在不转移原始检查记录的情况下，共同训练一个共享的基准模型。每个用户持有本地检查数据，并针对三个劣化方向转变（良好到轻微、良好到严重、轻微到严重）训练一个对数线性风险模型，考虑桥梁年龄、海岸线距离和桥面面积等协变量。通过小批量随机梯度下降进行本地优化。
基于FedAvg的连续时间马尔可夫链风险模型用于联邦桥梁劣化评估
📄 English Summary
FedAvg-Based CTMC Hazard Model for Federated Bridge Deterioration Assessment
This study proposes a federated framework for estimating a Continuous-Time Markov Chain (CTMC) hazard model for bridge deterioration, addressing the challenge of sensitive information in periodic inspection records that complicates cross-organizational data sharing under current governance constraints. The framework allows municipalities to collaboratively train a shared benchmark model without transferring raw inspection data. Each user maintains local inspection data and trains a log-linear hazard model focusing on three transition directions: Good to Minor, Good to Severe, and Minor to Severe, incorporating covariates such as bridge age, coastline distance, and deck area. Local optimization is conducted using mini-batch stochastic gradient descent.