矿山灾害感知数据是各类灾害过程实时的、客观的记录，科学分析感知数据的演变过程，通过聚类分析，从海量灾害感知数据中识别出有价值的灾害记录片断，对灾害防治具有重要意义。现有聚类模型无法表达数据演变过程的动态特征，也缺乏对演变过程相似性度量的方法，难以实现过程聚类。针对矿山灾害感知数据的特点和处理的实际要求，利用矿山灾害认知的领域知识来指导灾害感知数据演变过程的聚类分析，对原始感知数据进行数据约简、模式转换与特征提取，确定感知数据演变过程的关键事件和标识性特征，按照时间序列构建数据演变过程多层次、多尺度的特征链，分割灾害感知数据序列，研究序列的相似性度量方法，综合研究适用的聚类模型，形成灾害感知数据演变过程聚类分析的理论与技术体系。通过聚类后的灾害簇，分析矿山灾害的时、空演变规律，优化特征链，实现灾害过程的智能识别与自动、半自动检索。开发矿山灾害实时监测与预警系统，直接服务于矿山安全生产。

The perception data of mining disasters is real-time and objective recording during various mining disasters. Analyzing the evolution process of the perception data scientifically and identifying valuable fragments of disaster recordings from massive perception data by clustering method is very significant for disaster prevention. The current clustering models can't present the dynamic characteristics of the evolution process of perception data and lack the methods for measuring the similarity of evolution process; thus inapplicable for the process clustering. According to the characteristics of the perception data of mining disasters and the actual requirements for data processing, the proposed research will use the domain knowledge for cognizing mining disasters to guide the clustering analysis of evolution processes of disaster perception data, performing data reduction, mode conversion and feature extraction which on the original perception data. It then identifies key events and characteristic features of the evolution process of the disaster perception data, builds the multi-level and multi-scale feature chains of perception data evolution process in time-series, divides the perception data of disasters into segments, measures the similarity among the segments, studies the applicable clustering models for the evolution process of disasters perception data comprehensively, and develops the theory and technical system for clustering analysis of the evolution process of the disaster perception data. Moreover, through the clustering results, this research will analyze the spatiotemporal evolution rules of the mining disasters, optimize the feature chains, recognize the segments of disaster process from the perception data intelligently and retrieve them automatically or semi-automatically. Finally, a real-time monitoring and pre-warning system, serving for the mining safety directly, will be investigated and developed.