基于大数据的群体性突发事件预测研究

Based on big data analysis and the event library provided by the State Council Office of Emergency, a dynamic and open-sourced database about massive emergencies or collective actions of China will be built accordingly. It helps to describe and therefore figure out the attributions and regularities of collective actions, which make it possible to solve the relationships of related factors and estimate their coefficients. Then, we can predict collective action dynamically and provide precautions of them. The core issue concerns us is how to built the predicting model of collective actions, which will be resolved by the team via two channels: (a) the team will calculate, estimate and predict parameter values of individual and group actions using multiple methods of experiments, such as field experiments, big data monitoring, simulations, internet-based experiments, etc.; (b) the building of prediction model is conducted by combining a serial of analytical technologies, such as big data analysis, machine learning, correlation analysis, neural network analysis, clustering and discrimination skills and propensity scoring matching. This project shows a strong creativity and novelty in terms of: (a) it takes the lead in predicting collective actions in China, and the model is constructed by big data analysis, computer simulation, machine learning and other methods; (b) the prediction process is dynamic, which means that parameter values of related factors and variables will be updated according to instant changes of events and situations. The dynamic prediction mechanism overcomes the problem of lateness and weakness of prediction; (c) the combined big data prediction, including societal aspects, economic development, cultural tendencies, building and climate information and other methods, will be utilized to improve the accuracy and density of the prediction and precaution. Fully supported by the national basic research plan (973 program) that provides necessary working conditions and research foundations, this project aims and hopes to make contributions to improving capabilities of China’s emergency management.

项目基于大数据挖掘技术与国务院应急办的数据资料，建构群体性突发事件动态开源数据库。全域刻画其属性特征、时空规律与演化路径，解析影响因素的函数形式并估计相关变量参数，实现事件动态预测与分级预警。关键科学问题是建立预测模型，拟从两方面解决:（1）综合多种实验手段。通过真人场景实验、大数据监测、仿真模拟、互联网实验等获得个体行为与群体行为的参数规律;（2）结合多种分析技术。通过大数据分析、机器学习、关联分析、贝叶斯估计、神经网络分析、聚类分析、倾向值分析等建立预测模型。项目创新之处包括:（1）基于大数据分析、机器学习与仿真模拟，率先建立事件预测模型;（2）采用动态参数估计方法，模型参数伴随事态发展持续更新优化，解决预测滞后问题;（3）结合地理、经济、社会、气象、建筑、水文等高维度信息进行大数据组合预测，提高预测精度与密度。项目依托973团队的研究基础和工作条件，期待为我国应急管理事业做出贡献。

人类群体行为是群体突发事件的科学内核，项目基于社会计算研究人类群体行为。包括: (1)人类群体行为生命周期模型。人类群体行为宏观涌现生命周期规律，必然存在微观行为支撑。团队提出生命周期模型，并通过数学建模、仿真模拟、真实案例三重验证。在宏观涌现与微观行为之间，打通诸多层次、诸多环节，实现大尺度、全谱系研究; (2)人类群体行为与公共安全动力学仿真。人类群体行为规律用于公共安全领域，发现紧急状态下人类行为模式与粒子运动无差异。基于社会物理学，采用微观粒子行为模拟人群动力学。重点研究恐怖砍杀机制、人群踩踏机制、见义勇为机制、广场疏散动力学机制; (3)人类网络群体行为生命周期研究。人类网络群体行为与人类群体行为紧密联系。一是求解人类群体行为现实-虚拟场域之间的转换机制，团队进行了客观实证测量。二是人类网络群体行为涌现规律发现，采用四个参数预测网络舆情全周期过程; (4)人类网络群体行为动力学微观ABM仿真。人类网络群体行为的宏观涌现，需要微观行为机制。使用ABM多主体建模仿真，动态还原个体行为汇聚成宏观涌现的全过程; (5)疫情期间科研与应急管理研究。基于人类群体行为成果与ABM仿真方法，提交疫情预测专报。有效支撑决策、服务疫情防大局。项目超额完成计划任务: (1)论文发表。发表SCI/SSCI论文12篇，均为JCR一区（近3年最高），包括6篇中科院Top期刊论文。另外，发表CSSCI论文7篇、Science杂志eLetter文章5篇; (2)知识产权。申请发明专利2项、获批软件著作权15项; (3)算法与系统开发。开发网络舆情智能预测算法，获得大数据高科技公司商用，获得市场价值转化;(4)人才培养方面。晋升教授1名、副教授1名，培养2名博士生、硕士生6名、本科生4名。依托该项目，负责人成长为国内社会计算领域青年学者代表。入选2018湖南省“湖湘青年英才”，入围2018万人青拔面试阶段，入选2019国家民委优秀中青年专家，入选2020年度教育部青年长江学者。项目培育国家级重大项目2项。项目在人类群体行为、社会公共安全、网络舆情治理方面存在应用价值，其衍生成果将服务于国防军队建设。