Road Condition Perception Based On Trajectory Data Mining

In recent years,the development of road traffic has greatly facilitated people’s life,but the problems that followed were the difficulty of perception of traffic conditions,such as early warning of road conditions,forecast of coming congestion,etc.With the development of computer technology,the Intelligent Transportation System(ITS)combined with the transportation provides us with more and more solutions to perceive road conditions.In Intelligent Transportation System,a variety of data will be generated,of which the trajectory data can provide a large amount of information for mining.If we can analyze the collected trajectory data with some other data such as the data from sensors capturing the current road traffic flow,and the driver’s smartphone capturing acceleration and GPS data,we can make better analysis on the data to improve the accurate and timeliness of traffic conditions perception.In order to solve the problem of road condition detection,the paper demonstrates a road condition detection approach based on crowdsensing.The embed acceleration sensor of mobile phone are applied to excellently detect the vibration situation.Thus,the acceleration status and GPS data can be collected from running vehicles,representing the road condition and traveling tracks.After data collection,learning algorithms are used to deduce the actual road condition.This paper proposes a double-clustering based approach to implement road environment monitoring based on crowdsensing.The first level clusters the accelerating feature matrix to find out the corresponding coordinate category,and then the second level clustering is performed on the GPS coordinates to deduce the actual road condition.Finally,the moderate results obtained from on-road experiments prove the correctness and efficiency of our approach.In order to solve the problem of traffic flow prediction,the paper demonstrates a traffic flow prediction method based on temporal and spatial attention,and proposes a gated temporal and spatial attention extraction model GSTAN,which innovatively applies the Transformer model to the field of traffic flow prediction and targets the model,the structure is modified to extract spatial attention and temporal attention respectively to capture spatial dependence and temporal dependence.Then gated fusion is applied to fuse the temporal attention and spatial attention.In order to capture the periodic information of the traffic flow,the traffic flow in the previous hour and the traffic flow at the same time a day ago and the traffic flow at the same time a week ago are used as input to calculate the impact of the traffic flow in different time periods on the future time flow.After that,the three-period forecast data is weighted to obtain the final forecast result.On the METR-LA and PEMS-BAY data sets,the GSTAN model has excellent performance on the traffic prediction task of fifteen minutes prediction,half an hour prediction,and one hour prediction.The paper has 31 figures,5 tables,and 76 references.