基于大数据的航天发射系统安全性实时评估方法

Safety is a primary requirement of Space Launch System which has the periodic intermittent characteristic of short-term operation and long-term shutdown since the success or failure of launch mission is directly influenced by minor defects or incipient faults during the operating process. This project deals with the theories and technologies of its operational safety, including accident evolution modeling, monitoring and early alarming of operating condition as well as real-time safety evaluation and prediction. Aiming at establishing theories and methodologies of real-time analysis and evaluation of operational safety for the Space Launch System, four major aspects considered are as follows: 1) model the evolutionary mechanism of safety accident under the condition of equipment fault, maloperation and process parameter overrun; 2) explore the big data analysis based technologies of abnormal operating condition monitoring and alarming including correlative relationship between operating conditions and process variables, modeling of operating condition as well as identification and early warning of abnormal condition; 3) investigate the operational safety of Space Launch System under the condition such as process parameter overrun, component fault and maloperation with man-in-the-loop; 4) establish index based real-time evaluation methods for operational safety including the evaluation index system, quantification model of safety index and system safety level. Meanwhile, the validation tests on typical subsystems of real launch systems will be implemented to reveal the effectiveness of above theories and methods. Combined the theoretical research and actual requirements of Space Launch System, this project is of both theoretical importance and practical significance for space launch mission.

航天发射系统具有短期运行和长期停用的阶段性间歇运行特点，各子系统运行异常、微小缺陷或故障就可能导致运行事故，直接影响发射任务的成败，其高风险决定了安全性是航天发射任务的首要要求。本项目开展航天发射系统事故演化建模、运行工况监测与预警、运行安全性实时评估与预测的理论和方法研究。重点解决工艺参数超限、故障和误操作下的安全事故演化机理，系统在大数据下的运行工况与过程变量关联关系、工况工作区建模、异常工况识别和预警，动态系统在工艺参数超限、设备/系统故障、人在回路误操作等情况下的运行安全分析，动态评价指标体系构建和运行安全性实时评估模型等问题，形成一套针对航天发射系统运行安全性的在线分析与实时评估理论和方法。本课题的理论研究与航天发射系统的实际需求紧密结合，在航天发射场典型系统进行实验和应用验证。本项目不仅具有重要的理论意义，对我国航天发射系统的安全性具有应用价值。

航天发射系统中异常状态或微小故障可能导致安全事故，其高风险决定了安全性是航天发射任务最根本的要求。项目开展了航天发射系统事故演化建模、大数据下运行工况异常监测与预警、运行安全性实时评估与预测的理论和方法研究。.1、航天发射系统运行事故演化模型。航天发射系统庞大相互耦合，航天发射系统事故演化规律和机理难以进行动力学建模。结合航天发射系统运行机理和知识，以系统功能为切入点，按照子系统、设备、危险因素等进行分解和本体抽取，建立了基于航天发射系统“危险因素-安全”知识图谱与运行数据结合的运行事故链式演化模型。.2、基于大数据的系统异常工况在线识别与预警方法。航天发射系统运行过程具有多模态和多阶段性，以样本的流形结构为切入点，研究过程运行特征提取与降维表示，提出了基于判别分析的流形保存投影算法和不平衡样本下基于代价敏感自编码网络的在线识别诊断算法，实现了动态过程监测和预警。.3、危险因素下动态过程运行安全分析与预测。航天发射系统监测数据呈现高维、非线性、噪声干扰等特性，项目提出了一种基于分数阶阵发混沌系统的微弱故障检测方法，可检测微弱故障。提出了无监督框架下的非线性特征提取，采用字典学习和栈式自编码网络实现航天发射系统运行监测数据的特征提取和故障诊断。通过系统运行安全性判决和运行安全域，实现航天发射系统运行安全预测。.4、航天发射系统运行安全性实时评估。提出了动态系统安全性判据，设计了指数形式障碍函数约束条件，给出系统运行安全域，形成面向该类连续非线性动态系统的安全性分析方法。以低温加注系统为对象，构建基于功能层级的系统运行评价指标集和要素集，结合事故链式演化模型和安全性判据，对危险过程运行安全性进行评估。.5、大数据下的运行安全性实时评估方法的应用验证。在某航天发射场进行了低温推进剂加注和发射飞行安全辅助决策的模型和算法验证，结果表明运行监控、故障诊断、辅助决策、安全分析预测和评估结果正确。