Identification and Prediction of Large Crowd Gatherings based on Transportation Big Data

Publication
Dissertation

In recent years, with the rapid development of China’s economy and the rapid increase of urban population, the imbalance between supply and demand becomes more notable as various commercial and entertainment activities attract large crowds in urban areas. Although large-scale activities can bring substantial economic benefits to the city and satisfy the spiritual needs of the people, they will also bring tremendous pressure on the transportation facilities around the activity area, leading to negative impacts such as chaotic safety management, traffic congestion. Stampede accidents, resulting in casualties, may occur from time to time. And the places where large crowds gather often have limited evacuation routes. When the density of crowds in these places exceeds the threshold, it becomes more difficult to evacuate the crowds. Therefore, prevention is more effective. This dissertation starts from exploring the problem of how people gather in the first place, and looks for key indicators and methods for early warning and prediction of large-scale crowd gathering. Using data-driven methods to extract urban dynamic travel information from urban multimodal traffic data. An anomalous mobility network model was established to characterize the crowd aggregation model. Traffic data and signaling data were merged to estimate real-time travel demand and population distribution. The main research contents are as follows:

(1) Mobility analysis for large scale crowd gatherings: This dissertation obtains massive mobility information with high spatiotemporal resolution from billions of subway smart card data, taxi GPS data, and bus GPS data, and identifies 14 large-scale crowd gathering events in Shenzhen. A crowd density evaluation method is proposed. It is found that the crowd gathering pattern is hidden in the mass travel pattern, which is difficult to be identified by the traditional travel characteristics analysis method.

(2) Modeling and early warning for large crowd gathering: This dissertation is based on the research paradigm of complex networks, using traffic zones as nodes, travel between traffic zones as edges, and analyzing mobility pattern from the perspective of the network. Using an information theory indicator, an anomalous mobility network model is constructed to identify essential mobility patterns during large crowd gatherings. Through the dynamic evolution of the structure of the anomalous mobility network, the early-warning parameters that are sensitive to the clustering situation of the crowd are found, and the threshold of the early-warning parameters is calibrated with the DBSCAN method. The model can rise early warning several hours in advance, and the crowd density at the time of early warning is only 30% of the peak density.

(3) Theoretical model of anomalous mobility network: Anomalous mobility networks can accumulate evidence for potential crowd gatherings using low-intensity connections, that is not the volume of traffic, but the diversity of traffic sources, can predict large crowd gatherings at early stages, which is a phenomenon of “little drops of water make a mighty ocean”. Then a theoretical model is established to proof the two insights of the anomalous mobility network.

(4) Dynamic travel demand estimation and population distribution prediction: This dissertation combines the historical data of mobile phone signaling and real-time traffic data to establish a dynamic estimation model of urban residents’ travel. The model takes full advantage of different data sources. Through this model, the population distribution of each traffic zone in the city is deduced. By extracting the spatiotemporal features with the recursive feature elimination method, a prediction model was established, which can predict population distribution 1 hour in advance. The relative error of the prediction is about 14%. When large crowd gathering occur, this model can predict the trend of population.

Advisor: Prof. Pu Wang

Defense committee: Prof. Zhongxiang Huang, Prof. Xiamiao Li, Prof. Kejun Long, Prof. Hui Liu, Prof. Lianbo Deng