With the widespread use of information technologies such as IoT and big data in the transportation business, traditional passenger transportation has begun to transition and upgrade into intelligent transportation, providing passengers with a better riding experience. Giving precise bus arrival times is a critical link in achieving urban intelligent transportation. As a result, a mixed model-based bus arrival time prediction model (RHMX) was suggested in this work, which could dynamically forecast bus arrival time based on the input data. First, two sub-models were created: bus station stopping time prediction and interstation running time prediction. The former predicted the stopping time of a running bus at each downstream station in an iterative manner, while the latter projected its running time on each downstream road segment (stations as the break points). Using the two models, a group of time series data on interstation running time and bus station stopping time may be predicted. Following that, the time series data from the two sub-models was fused using long short-term memory (LSTM) to generate an approximate bus arrival time. Finally, using Kalman filtering, the LSTM prediction results were dynamically updated in order to eliminate the influence of aberrant data on the anticipated value and obtain a more precise bus arrival time. The experimental findings showed that the suggested model's accuracy and stability were both improved by 35% and 17%, respectively, over AutoNavi and Baidu.
RHMX: Bus arrival time prediction via mixed model
