| dc.description.abstract | With the economic development, the existing road supply is not enough to load the rapid growth of traffic, making traffic congestion a common phenomenon in various cities. However, road supply can’t keep up with the growth rate of transport vehicles, and transportation control has become an effective way to reduce traffic congestion. If we provide accurate travel time estimates to drivers, drivers can choose substitute route to achieve traffic flow distribution and reduce traffic congestion. However, studies on travel time prediction are mostly applied to expressways and few studies related to travel time prediction for urban roads in domestic.
This study proposes a urban road travel time prediction method that use data mining techniques combined with eTag vehicle detectors. By using Weka data e mining software to test nearest neighbor method, support vector regression, multi-layer perceptron, regression tree and random forest etc. algorithms for experimental analysis. Among them, in the past related research, the traffic flow accumulation factor mostly considered only the short-term traffic flow, in this study, we found that if the traffic flow accumulation interval is greatly lengthened, the prediction accuracy can be greatly improved. In addition, this study also found that the traffic volume of large vehicles does have a significant impact on the travel time prediction. This study uses the eTag vehicle detector of the Taoyuan City Government Department of Transportation as the data source, and predicts the travel time of peak hours on weekdays and weekends of the south bound Taoyuan Chunri Road. Based on the actual measurement results, the prediction performance of this research method is a high-precision prediction in the MAPE evaluation standard, which can be effectively applied to urban road vehicle diversion management strategies. | en_US |