基于动态信息谱的流程工业事故致因信息转移机理研究

With the large-scale application of advanced control technologies and intelligent equipment in modern process industry, accident evolution mechanism is becoming so complex that traditional causing process model in the form of graph theories, which is developed on the basis of analytical model and statistics, cannot meet the needs of work safety in the future. Complex network and information theory provide a new approach to deal with accident-causing complexity. In the sense of Wiener-Granger causality, our research is to study the computational method for different types of information transfer among accident factors, and characterize the causing relation with these information transfer, and then construct the accident-causing complex network for process industry. On the basis of this network, the information transfer matrix among accident factors will be established, and the dynamic information spectrum describing the holistic feature of the information transfer among accident factors will be proposed. Taking advantage of the sensitivity of dynamic information spectrum to the key changes in the system, predictions in the process industry can be carried out with time extrapolation. Breaking through the limitation of reduction and analysis on the material flow and energy flow in accident-causing research in process industry, this research provides a new approach, i.e. the information theoretical method, for studying the accident evolution mechanism in process industry, which is of great theoretical significance.

随着先进控制技术及智能设备在现代流程工业中的大规模应用，流程工业事故演化机理日益复杂，基于解析模型和传统统计所建立起来的图论形式的事故致因模型难以满足流程工业未来安全生产的需求。复杂网络和信息理论为探索致因系统的复杂性提供了新途径。本研究在Wiener-Granger因果关系框架下，研究事故因素之间分类别信息转移计算方法，提出通过变量之间的信息转移来表征事故因素之间的致因过程，从而建立起事故致因复杂网络；在此基础上，构建事故因素信息转移矩阵，研究能用以刻画信息转移全局特征的动态信息谱，利用动态信息谱对于系统关键变化所具有的敏感指示特性，通过谱轨迹的时间外推，预判流程工业系统所产生的关键变化和事故的演化发生。本项目突破了流程工业事故致因研究中仅仅对物料流和能量流进行还原解析的限制，从信息转移的角度为流程工业事故演化规律的研究提供了新途径，对于流程工业事故复杂致因过程的研究具有重大理论意义。

随着先进控制技术及智能设备在现代流程工业中的大规模应用，流程工业事故演化机理日益复杂，基于解析模型和传统统计所建立起来的图论形式的事故致因模型难以满足流程工业未来安全生产的需求。复杂网络和信息理论为探索致因系统的复杂性提供了新途径。过去三年，本研究从复杂系统理论和信息理论的角度，就流程工业事故致因开展了以下研究：①提出了基于符号转移熵谱和Granger因果关系的综合网络推断STESGC方法，从多元时间序列推断加权有向关联网络。该方法解决了网络推断中的五个问题，即时间序列的连续性、强关系识别、关系量化、方向识别和时间延迟关系识别；②提出了一种基于致因网络的系统总体工况异常检测AN-SVM方法。该方法基于STESGC方法所推断出的致因网络，通过复杂网络理论计算拓扑特征指标，得到系统的结构表征，进而通过SVM模型的学习来研判系统是否发生异常；③提出了基于信息熵变的致因网络节点重要性排序方法。从节点移除带来致因网络信息熵变的角度，来量化事故致因的重要程度。所提出的方法分别在仿真系统（线性系统、一般非线性系统和耦合Rossler系统等）和真实的过程工业系统中得到了验证。