A Bus Arrival Time Prediction Method Based on Position Calibration and LSTM

Abstract

Bus arrival time prediction not only provides convenience for passengers, but also helps to improve the efficiency of intelligent transportation system. Unfortunately, the low precision of bus-mounted GPS system, lack of real-time traffic information and poor performance of prediction model lead to low estimation accuracy-greatly influence bus service performance. Hence, in this paper, a GPS calibration method is put forward, while projection rules of specific road shapes are discussed. Moreover, two traffic factors, travel factor and dwelling factor, are defined to express real-time traffic state. Then, considering both historic data and real-time traffic condition, a hybrid dynamic BAT prediction factor, which achieves accuracy enhancement by taking into account traffic flow evaluation results and GPS position calibration, is defined. A LSTM training model is construct to realize BAT prediction. Experiment results demonstrate that our technique can provide a higher level of accuracy compared to methods based on traditional time-ofarrival techniques, especially in the accuracy of multi-stops BAT prediction. INDEX TERMS Bus arrival time prediction, LSTM model, GPS data calibration.

A review of bus arrival time prediction using artificial intelligence

“…The first important and crucial step of designing a BAT prediction system is selection of the suitable model. These models based on their characteristics and application in field of BAT prediction are mainly divided into four categories: probabilistic models (Abkowitz et al, 1987; Anderson & Goodman, 1957; Dhivyabharathi et al, 2016; Guenthner & Hamat, 1988; Hans et al, 2015a, 2015b; Krbálek & Seba, 2000; Lee et al, 1968; Lin & Bertini, 2004; Tian et al, 2018), historical models (Biagioni et al, 2011; Maiti et al, 2014; Wepulanon et al, 2017), statistical models (Chen et al, 2007; Huang et al, 2021; Patnaik et al, 2004; Sinn et al, 2012; Xiang et al, 2020), shallow machine learning models (Bin et al, 2006; Chen, 2018; Chen et al, 2007; Chien et al, 2002; Fauzan et al, 2019; Hua et al, 2017; Huang et al, 2021; Jalaney & Ganesh, 2020; Ji et al, 2016; Kalaputapu & Demetsky, 1995; Kee et al, 2017; Khamparia & Choudhary, 2019; Lai et al, 2020; Lam et al, 2019; Li, 2017; Lin et al, 2013; Peng et al, 2018; Treethidtaphat et al, 2017; Wang et al, 2014; Yang et al, 2016; Yin et al, 2017; Yu et al, 2010, 2011; Zhang et al, 2017), and deep machine learning models (Agafonov & Yumaganov, 2019; Alam et al, 2020; Han et al, 2020; Huang et al, 2019; Kalaputapu & Demetsky, 1995; Lingqiu et al, 2019; Liu, Sun, & Wang, 2020; Liu, Xu, et al, 2020; Pang et al, 2019; Panovski & Zaharia, 2020; Pa...…”

“…It has been observed that almost all the studies involved in BAT prediction are for the next stop only. Recently, different deep machine learning based models were employed for multistep BAT prediction which outperformed all other state of the art models (Alam et al, 2020; Han et al, 2020; Pang et al, 2019). Although, it was found that reported accuracy was not as accurate as compared with one and two step prediction.…”

“…Alam et al (2020) also investigated multistep ahead arrival time of bus with weather effects. Two variants of LSTM weighted and unweighted were also constructed to realize multistops bus prediction which considered real‐time traffic, historic trajectories, and bus stop dwell time (Han et al, 2020).…”

“…It has become clear that ANN and deep learning based models are most popular in the field of BAT prediction. There are various studies which have established that these models outperform almost every method discussed previously (Do et al, 2018;Han et al, 2020;Maiti et al, 2014). Next section provides a detailed review on the application of AI based models in this field (Table 2).…”

Multistep Prediction of Bus Arrival Time with the Recurrent Neural Network

“…Several studies adopted non-RNN networks for predicting bus arrival and operation times using (1) MapReduce-based clustering with K-means [4], (2) a backpropagation (BP) neural network model [5], (3) a particle swarm algorithm [6], (4) a wide-depth recursive (WDR) learning model [7], and (5) RNN with the time series such as long short-term memory (LSTM) [8]. Models with LSTM processed the historical data of the global position system (GPS) and bus stop locations with the influence of different routes, drivers, weather conditions, time distribution [9], heterogeneous traffic flow, and real-time data [10][11][12]. e temporal and spatial RNN network with ConvLSTM or a spatiotemporal property model (STPM) was originally used to predict the precipitation [13].…”