REAL-TIME SHARED-MOBILITY FLEET INTELLIGENCE FOR VEHICLE AVAILABILITY, TRIP DEMAND, UTILIZATION AND BATTERY-AWARE REBALANCING - (MOVE GRID)

Abstract

Shared electric bicycles and scooters have become a common way of covering short distances in cities, especially between metro stations, offices, colleges, and residential areas. The service depends on a simple promise: a charged vehicle should be available close to where a rider wants to start a trip. In practice, vehicles collect at popular destinations, run out of charge in busy zones, and sit idle in quiet areas, while riders elsewhere open the app and find nothing nearby. Operators usually respond with fixed rebalancing routes and battery swap rounds planned from experience. This paper presents Move Grid, a real-time fleet intelligence system that uses streaming vehicle and trip data to forecast demand, monitor availability and utilisation, and plan batteryaware rebalancing. Each vehicle reports its GPS position, battery state of charge, lock status, and fault codes every thirty seconds, and the booking platform records trip starts, trip ends, and failed searches where no vehicle was found. These streams are published to Apache Kafka and processed with Apache Flink, which maps positions to H3 hexagonal zones, validates readings, and computes rolling counts of available vehicles, usable vehicles above a battery threshold, and open demand for every zone. The processed data is stored in a TimescaleDB time-series database, while a daily batch pipeline prepares training data that combines trip history with weather, holidays, and local events. The analytical core consists of three models. A LightGBM demand model forecasts trip starts per zone for each of the next four thirty-minute intervals using lagged demand, time of day, day type, rainfall, and nearby activity. A battery depletion model estimates the energy each vehicle will use on a typical trip in its zone and predicts when it will fall below the rentable threshold. A rebalancing optimiser formulated as a mixed-integer program with Google OR-Tools decides how many vehicles to move between zones and which low-battery vehicles to prioritise for swapping, subject to the capacity and shift time of each service van. The system is designed for fleet operators, city mobility teams, and field supervisors. It recommends moves and swaps but leaves dispatch decisions to the operations staff, and it respects parking zones and restricted areas defined by the city. Future work includes pricing incentives that encourage riders to end trips in under-supplied zones, integration with public transport timetables, and support for mixed fleets of bicycles, scooters, and mopeds.