HIPE, the Hybrid Intelligence Planning Engine, combines machine learning with the knowledge of experienced transport planners to plan timetables for bus, tram and metro. We built the first version with EBS and then extended it with RET and HTM, within the InnOVatieroute partnership of EBS, RET, HTM and MRDH.
In the first phase we developed a working model on ten EBS bus lines in four months. Where the old planning method only looked at the past, HIPE predicts travel time based on traffic, weather, season and passenger demand. That delivered roughly 16% better punctuality and a five times faster planning process.
HIPE does not only give a prediction, it also shows where that prediction comes from. The transport planner decides, the model calculates and substantiates. That way the organisation can check and explain a choice, instead of blindly following a system.
In the further development we made the data pipeline real-time, investigated a more powerful prediction model and connected HIPE to the Hastus planning software. The model is now numerically validated for EBS, RET and HTM, with error margins at the level of seconds. Users rated their trust in and the explainability of the predictions a 7.7 out of 10.
The next step is a period of running in shadow mode against the current planning method, tuned further per operator. With that, HIPE is on its way to daily use.
