Forecasts

Forecasts – An Intelligent Look into the Future

Based on our real-time data, DB Lightgate generates AI-supported forecasts for occupancy development along a journey.

Each time one of our sensors measures the occupancy of a passing train, DB Lightgate predicts the occupancy by car for the next following stations. Therefore, the forecast is always based on the current real-time data and achieves the highest possible quality.

This delivers two main benefits:

Passengers at the platform receive early information on where to expect available seats on the train.
Operations gain valuable time windows for decision-making—e.g., targeted dispatching at junctions or activation of additional services.

Our forecast models are based on historical movement patterns, real-time occupancy, and the dynamics of individual lines. They are continuously trained, tested, and refined to enable longer forecast horizons while maintaining high accuracy.

The forecast is so accurate that it seamlessly transitions into real-time data. Each calculated forecast is verified shortly afterward against the measured real-time data. This way, the quality is monitored around the clock.

The forecasting engine calculates the change in occupancy per carriage for the next five stations. The station, the train type, the carriage position, and the time are important variables for the prediction.

Dispatchers and train traffic controllers know not only the real-time occupancy but also the predicted load for the upcoming stations. This improves decision-making for stable operations.

Our passengers can position themselves at the platform in good time thanks to the carriage-specific ultra-short-term forecast. Even without real-time data, the occupancy of the next train can be provided.

Differentiated Models for Different Requirements

Requirements in metropolitan transport differ significantly from regional services. While metro systems are characterized by high frequency and short dwell times, regional transport focuses on line patterns, transfers, and longer travel times. Our models are correspondingly differentiated and modularly expandable.