基于多源异构数据的学生群体性异常行为预测模型研究与优化

如何利用前沿的深度学习技术对高校大数据中的学生群体性异常行为进行预测，是教育大数据研究领域中的一个重点与难点。本课题将校园大数据中的传统的一卡通等数据扩展到校园监控数据、通知新闻数据等，具有一定的挑战性。课题主要研究内容如下：（1）通过YOLOV3图像检测和卷积神经网络CNN技术，以校园时空数据为主，融合多源数据，构建基于ConvRes的校园学生人群流量预测模型，以实现非正常集会的挖掘。（2）提出一种基于N-BiLSTM的嵌套模型来对文本特征进行特征表示并融合到预测模型中，以考虑到高校相关部门的新闻通知信息对校园的人群流量分布造成一定的影响；（3）提出一种基于CNN的结合权重共享策略的稀疏连接方法来对预测模型进行压缩，以加速模型在实际预测过程中的时效性。本课题通过校园视频等强特征性数据来深入挖掘学生的群体性异常行为，课题的研究对加强校园维稳以及对大学生行为管理等有重要的现实意义。

How to use cutting-edge deep learning technology to predict the group abnormal behavior of students in big data in colleges and universities is an important and difficult point in the field of educational big data research. This subject is challenging to extend the traditional one-card data in campus big data to campus monitoring data and notification news data. The main research contents of the topic are as follows: (1) Based on YOLOV3 image detection and convolution neural network (CNN) technology, the campus spatiotemporal data is used as the main data, and multi-source data is integrated to build a prediction model of the flow of students on campus based on ConvRes, so as to realize the mining of abnormal assembly. ; (2) A nested model based on N-BiLSTM is used to characterize the text features and integrate them into the prediction model to take into account that the news notification information of the relevant departments of the university has a certain impact on the distribution of campus crowds; (3 ) A sparse connection method based on CNN combined with weight sharing strategy is used to compress the prediction model to accelerate the timeliness of the model in the actual prediction process. This topic uses campus video and other strong characteristic data to dig deep into the group abnormal behavior of students. The research of this topic has important practical significance for strengthening the stability of campus and the behavior management of college students.