Sun Qingheng
North China University of Water Resources and Electric Power, Zhengzhou, 450000, China.
Sci Rep. 2025 Aug 12;15(1):29502. doi: 10.1038/s41598-025-15555-y.
As China's aging problem intensifies and urban-rural resource allocation remains uneven, the shortage of property-based elderly care service resources in counties within urban-rural integration zones has become increasingly prominent. Traditional early warning models struggle to dynamically capture the supply-demand fluctuation patterns of property-based elderly care resources. To optimize the "property + elderly care" service, this paper proposes an innovative dynamic early warning model based on the Long Short-Term Memory with Attention Mechanism (LSTM-AM), introducing "property response" and "service supply-demand spatio-temporal coupling degree" as dual core indicators and establishing a dynamic threshold-based graded early warning mechanism. This method shifts from reactive remediation to proactive adaptation, offering intelligent tools to support the coordinated development of property-based elderly care services in urban-rural integration areas. Results show that, taking Xuchang City, Henan Province as an example, by integrating multi-source time-series data such as county-level population aging rates, property-based elderly care service coverage rates, community facility maintenance data, and resident demand feedback, the model's prediction error (MAE ≤ 0. 12, RMSE ≤ 0. 18) was reduced by 35% compared to traditional methods, and the warning accuracy (F1-score = 0. 89) was significantly superior to traditional models. The model can predict resource shortages 8-12 months in advance and generate facility allocation and personnel scheduling plans.
随着中国老龄化问题加剧，城乡资源分配不均衡，城乡融合区内县域基于物业的养老服务资源短缺问题日益突出。传统预警模型难以动态捕捉基于物业的养老资源供需波动模式。为优化“物业 + 养老”服务，本文提出一种基于带注意力机制的长短期记忆网络（LSTM - AM）的创新动态预警模型，引入“物业响应度”和“服务供需时空耦合度”作为双核指标，并建立基于动态阈值的分级预警机制。该方法从被动补救转向主动适应，为支持城乡融合区基于物业的养老服务协同发展提供智能工具。结果表明，以河南省许昌市为例，通过整合县级人口老龄化率、基于物业的养老服务覆盖率、社区设施维护数据和居民需求反馈等多源时间序列数据，该模型的预测误差（MAE≤0.12，RMSE≤0.18）相比传统方法降低了35%，预警准确率（F1分数 = 0.89）显著优于传统模型。该模型可提前8 - 12个月预测资源短缺情况，并生成设施配置和人员调度计划。
