Predictive Analytics of Time Series for Preventive Maintenance of Urban Infrastructure: A Research Plan and Experimental Design Based on MetroPT-3

This article-style preprint presents a research plan for predictive analytics of time series in preventive maintenance of urban infrastructure. The work motivates the research gap between general predictive maintenance research and urban infrastructure needs, selects MetroPT-3 as a reproducible urban-transport predictive maintenance benchmark, and proposes an experimental protocol for comparing baseline, statistical, deep learning, Transformer-based, and hybrid time-series models under chronological validation. The preprint consolidates the current stage of a master's research project, including problem formulation, dataset strategy, preliminary data understanding, preprocessing design, model classes, evaluation metrics, limitations, and a publication plan. The planned evaluation emphasizes forecasting accuracy, anomaly detection quality, false-negative analysis, and warning lead time. The document is intended as a citable preliminary research output before the empirical model-training stage.