A Novel Traffic Flow Data Imputation Method for Traffic State Identification and Prediction Based on Spatio-Temporal Transportation Big Data

摘要

Traffic state identification and prediction are playing an important role in intelligent transportation systems (ITS) research, especially in the era of big data when a variety of big data resources are available. In this paper, two kinds of real-world collected spatiotemporal datasets are analyzed. These data sets were captured by microwave and video detectors from Haining City of China. First of all, we propose a novel method which transfer one dimensional time series data into two dimensional pattern for missing data compensation and data fusion. After that, we build a special kind of recurrent neural network (RNN) which is usually called long short term memory (LSTM) networks to train the time series traffic flow data sets. We compare the prediction accuracy between the imputed data and the original data. Experiments show that the prediction error has lower RMSE score by using proposed novel method for imputation than original data without using this method.