The Train Brain for bus

Important results from the project

The aim of the project has been to develop an AI-based real time computation service for bus traffic. The project has been successful - The Train Brain can create more reliable traffic information compared to existing forecasting solutions, new types of traffic management decision support and can be used as a tool for more robust timetables.

Expected long term effects

In the area of traffic information, the project has shown that delay forecasts from The Train Brain are 74% more accurate than the current forecasting service used , 1 minute before the trip. 7% of all buses in the SL system depart more than 2 minutes early. The Train Brain can reduce the risk of passengers missing their bus due to this, from 4% to 1% of all departures. The Train Brain´s long-term forecasts are more accurate 6 days in advance than SL´s existing forecasting service are 1 minute in advance.

Approach and implementation

The Train Brain model has been trained using a dataset with historical traffic data, current bus schedules, a real-time stream with check-ins at bus stops in the SL service. The trained model has been used to estimate all driving times in bus traffic. To determine the quality of delay forecasts, we have compared Train Brain forecasts with the forecast services used by SL today. This comparison has been made in many different dimensions, such as regularity, time stops, contracted bus area, forecast source and at different times before arrival and departure.