Twin Disruption Prediction Accuracy (TDPA)

GMI-RAIL-046Rail TransportationIntegration-RequiredPattern O-3: Twin Fidelity

Definition

Ratio of network digital twin disruption impact predictions to actual disruption outcomes, measuring whether the twin's simulated delay propagation, recovery time, and resource consumption match what happens in the real network under disruption.

The Measurement Gap

Why This Metric Didn't Exist Before

Network digital twins for railroads are new. The concept of simulating an entire rail network (trains, crew, track, signals, weather) and running disruption scenarios through it is a capability that didn't exist at operational scale until recently. Before twins, disruption planning relied on experience and playbooks: dispatchers knew from years of experience roughly how a major blockage would propagate, and they applied that intuition to recovery planning.

Digital twins promise to replace intuition with simulation. But simulation is only useful if it's accurate. The specific failure mode TDPA measures: twins model steady-state conditions well but consistently underestimate disruption impact because they don't model degraded-state constraints. When a disruption hits, crew are approaching Hours of Service limits, equipment is already deployed, and maintenance windows can't be easily moved. The twin that doesn't model these constraints predicts a 3-hour recovery that actually takes 8 hours.

Here's the decision TDPA supports: Can the network twin be trusted for disruption planning and recovery resource allocation? If TDPA is low with consistent underestimation bias, every disruption plan based on twin predictions should include a safety factor, and the twin's model scope needs to expand to include the degraded-state constraints it's currently ignoring.

Without TDPA, you discover the twin's inaccuracy during actual disruptions, when it's too late to allocate additional recovery resources. The dispatchers learn to distrust the twin and fall back to intuition, negating the investment.

Formula

Formula:

TDPA = 1 − |Predicted_impact − Actual_impact| / Actual_impact

Variable Definitions:

Variable Definition Unit Source Classification
Predicted_impact Twin's simulated total delay-minutes (or recovery hours, or resource consumption) for the disruption scenario Minutes/hours/dollars Network twin NATIVE
Actual_impact Actual total delay-minutes (or recovery hours, or resource consumption) from the same disruption Minutes/hours/dollars Operations system NATIVE

Sign Convention: 1.0 = perfect prediction. Lower = less accurate. Track directional bias separately: consistently underestimating impact (actual > predicted) means recovery plans are under-resourced.

Data Requirements:

Data Element Source System Classification
Twin disruption simulation outputs Network digital twin NATIVE
Actual disruption outcomes (delay, recovery time, resource cost) Operations system NATIVE
Disruption scenario parameters (location, duration, type) Operations / incident log NATIVE
Twin model scope documentation (what's in-model) Twin documentation NATIVE

Edge Cases and Measurement Boundaries:

Disruptions the twin wasn't designed to model present a scope limitation, not a model error. If the twin doesn't model crew HOS limits, equipment shortages, or maintenance-of-way window interactions, it will systematically underestimate disruptions that hit those constraints. Track which disruption factors are in-model and which aren't. The twin may be accurate for small disruptions (1-hour blockage) but inaccurate for large disruptions (8-hour blockage) because cascading effects compound nonlinearly: segment TDPA by disruption duration. Minimum sample: 10 disruptions per quarter, or retroactive simulation of historical disruptions.

Proposed formula, pending validation pilots.

At a Glance

Category
Passenger Experience and Safety Analytics Integrity
← Schedule Resilience Under Disruption Index (SRUDI)Alert Actionability Ratio (AAR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment