SCOUT: Scenario-Oriented Context-Gated Forecasting with Explicit Lower-Bound Projection for Eligible Railway Coal OD Flows

Abstract

Heterogeneity in route activity and the need to clearly distinguish among observations, scenario assumptions, and operational requirements pose challenges for the short-horizon forecasting of railway coal origin–destination (OD) flows. SCOUT implements overlapping design objectives for longer-term tendencies, periodic variation, and short-term changes through 15 explicitly defined temporal and constructed-proxy reference components. Their construction provides temporal meaning, while a context-dependent simplex gate adapts their combination. Validation-based calibration is followed by separate scenario adjustment and route-level lower-bound projection. The case study concerns one railway system and 210 routes retained after training-period screening of inactive and rare-activation routes, covering 98.42% of held-out network flow. Across 171 held-out forecast origins under a purged chronological split, the lower-bound-compliant forecast reduces RMSE relative to Recent-4 by 6.34%, 3.21%, and 2.89% in the intermittent, regular active-flow, and backbone strata. The directional comparison with Recent-4 persists under the reported temporal-boundary and threshold checks, while the modern benchmarks and architectural diagnostics show cluster-dependent trade-offs rather than a uniform benefit from every module. Projection eliminates violations of the specified lower bounds, but does not establish network-capacity feasibility. For intermittent routes, positive-event coverage of the descriptive residual band is only 4.73%. The evidence supports a structurally organized forecasting representation and an auditable adjustment interface; transfer to other systems and downstream operational benefits remain to be evaluated.