Urban road traffic speed estimation for missing probe vehicle data based on multiple linear regression model

GPS-equipped probe vehicles can collect reliable traffic speed information for real-time traffic state estimation in urban road network. However, there exist some road segments with missing or sparse probe vehicle data, which will reduce the accuracy and robustness of estimation. In this paper, we presented a spatial-temporal method based on multiple linear regression model to calculate the traffic speed of the segments without sensor data by fusing the information from adjacent interval time and road segments. Meanwhile, a heuristic method was designed for model parameters training with linear computational complex. It tries to make full use of the information from GPS data by selecting neighboring nodes with highest correlation coefficients dynamically, which will adjust the model parameters for different missing situations. The experiments on performance evaluation were carried out on real probe data from 2017 GPS-equipped taxis. The results show that the information from adjacent interval time and road segments is helpful for missing data estimation. Our model provides a decrease in root mean square error of 73.3% when compared to a baseline approach.