Using Loop Detector Big Data and Artificial Intelligence to Predict Road Network Congestion

Abstract

Understanding the temporal-spatial congestion evolution is important to mitigate traffic congestion and improve traffic efficiency. Most studies used floating car data to analyze the urban congestion, however, its market penetrate is low in many cities, thus the data is not enough in terms of quantity and coverage. Loop detector is the most frequently used sensor, its data has the attributes of long-term and large-scale coverage, and utilizing the loop detector big data is helpful to analyze the congestion evolution. Therefore, this study proposes a data-driven congestion analysis approach, which consists of loop detector data processing, traffic simulation, and artificial intelligence to predict the urban temporal-spatial congestion evolution. A case study in Tianjin, China is conducted, and the case study result shows that the evening peak has more serious traffic congestion than the morning peak, the prediction accuracy of feed forward back-propagation neural network (BPNN) increases with the time interval aggregation level increasing, and the prediction accuracy is 85.7% with 30 min interval aggregation.