Activity-travel Feature Extraction And Urban Crowd Density Prediction And Analysis Based On Mobile-phone Signaling Data

With the accelerating urbanization process in China,urban population and car ownership increase rapidly,leading to increasingly severe road traffic congestion and frequent traffic accidents,and public safety accidents.Urban crowd density prediction can assist traffic control and management,alleviate urban traffic congestion,and ensure urban public safety.Mobile-phone signaling data have many advantages such as realtime,complete spatial-temporal coverage,no extra cost,and contain rich activity-travel information,and are suitable for urban crowd density prediction.It is essential and practical to research extracting activity-travel features from mobile-phone signaling data and applying them to urban crowd density prediction.Because raw mobile-phone signaling data are large and noisy,this thesis firstly studies methods for mobile-phone signaling data preprocessing.The ray casting algorithm in computational geometry is used to preprocess the base station table,and then the mobile-phone signaling data in the research area is screened out according to the base station table.The data processing method based on time threshold and distance threshold is applied to remove the ping-pong handover data,outlier data,and repeated data,which are the three most common abnormal mobile-phone signaling data.The proposed data processing methods are all implemented in the form of SQL scripts and transformed into Map Reduce tasks with the help of the Hive platform for efficient execution in a distributed form.In order to extract the activity-travel features from mobile-phone signaling data,this thesis proposes a procedure for activity choice behavior identification,including identifying the stop points,identifying the home and work location of the users,and identifying activity location,activity type,and activity duration.This thesis defines the concept of activity-travel features in this study and proposes a method of constructing activity-travel features based on activity choice behavior information.This thesis establishes a spatial-temporal convolutional crowd density prediction model based on activity-travel features.The model uses the spatial-temporal convolutional module to consider the spatial-temporal correlation of crowd density and uses the attention-based feature fusion module to introduce the influence of activitytravel features and external factors(such as weather conditions and holidays)on crowd activity and crowd density.In addition,the proposed model can be used to predict the crowd density on irregular-shaped divisions.Taking the real-world mobile-phone signaling data from a district of Suzhou City as an example,the performance of the proposed model on the crowd density prediction task is tested.In addition,the long short-term memory neural network,and the spatiotemporal graph convolutional neural network,and other models are chosen for comparative experiments on the same dataset.The experimental results prove that the spatial-temporal convolutional model for crowd density prediction based on activitytravel features proposed in this study can consider the spatial-temporal correlation of data as well as the influence of crowd activities and external factors and achieve a better prediction accuracy of crowd density.Through error analysis,it is verified that the error of crowd density prediction is related to the activity and traveling of the urban crowds in different temporal and spatial scenarios.In the time or area where crowds travel more actively,such as the morning rush hour or the commercial district with a dense population,the error of crowd density prediction is also relatively significant.This thesis further analyzes the relationship between external factors and crowd activity and traveling and verifies that external factors indirectly affect the traveling of crowds and crowd density by affecting crowd activities.In addition,the influence of external factors on the traveling of urban crowds may be related to the activity choice behaviors(such as home activities,work activities,or other activities)and other travel-related factors(such as commuting distance).