Machine Learning-based Uptime-Prediction for Battery-friendly Passenger Information Displays

Personal Information Displays (PID) at bus stops help making the usage of public transport more attractive. If no electric grid is nearby, however, the installation of PIDs is very expensive due to the high wiring costs. To resolve this issue, the partners of the R&D project IoT-STOP develop a novel PID system that will be independent from the access to power lines. The system uses e-papers as displays that can be accessed using a cellular network. To prevent long, energyintensive idle listening, the network receiver operates only when the passenger information, in particular, the Expected Times of Arrival (ETA) of the buses, is updated. Between two updates, the receiver is switched off such that adjustments after sudden events are not possible. Therefore, the update periods have to be carefully selected. In this paper, we introduce a predictor that estimates time intervals between updates. Our method is based on linear regression using samples of previous bus rides to forecast arrival times. Its predictions are applied by an algorithm to detect areas during the journey of a bus at which its ETA at a later stop changes with a certain probability. The forecasted times for passing such areas are then selected to update the PID at this stop. In addition, we present a number of tests of the predictor carried out at some bus stops in Bergen, Norway. The results show that the proposed method indeed predicts sensible update times of the PID systems.