Research On Impact Of The Logging Frequency Of Cellphone Gps On The Extration Accuracy Of Travel Information

In the process of the development of the city, a series of "city disease" is produced, such as traffic congestion, traffic pollution. In order to master the traffic demand, to make traffic planning and design more scientific and reasonable and to guide the development of transportation system, urban transportation present situation and the operation mechanism must be deeply analyzed. The traditional resident trip survey has many problem of the accuracy, so it needs a new technology-cellphone GPS which can not only reduce the survey cost, also can improve the data quality. Although the existing research on cellphone GPS data is very thorough, and is concentrated on data mining, the time parameters in the process of data acquisition is an important factor affecting the quality of data, so further researches are needed.In this paper, impact of different cellphone GPS logging frequency on the extraction of traffic travel information is analyzed. Based on single factor method, the GPS data of different logging frequencies were obtained by the design of the data acquisition experiment and the data pretreatment process. According to the characteristics of data samples and research purpose, the instantaneous velocity is chosen as a feature vector and the traffic mode conversion points and traffic modes are identified by artificial neural network (ANN) and random forest algorithm (RF).Accuracy of different traffic mode conversion points are similar and with increase of the frequency, accuracy decreases. When sampling interval is 5 seconds, all traffic mode conversion points recall rates are more than 90%, and accuracy rates are more than 80% with the threshold value of 120 seconds. Considering the cost of survey and data, the sampling interval can be valued in 1-20 seconds in practice.At the same time, four traffic mode (pedestrian, bicycle, bus and car) are identified, ANN and RF have the same conclusions:when the sampling interval is 1 second in ANN or the sampling interval is 2 seconds in RF,non-maneuver accuracy is above 85% and maneuver accuracy is more than 80%; accuracy will increase with the logging frequency decreasing. And for the same purpose, the sampling interval is in the range of 1 to 10 seconds.