不确定环境下考虑时变因素航空安全多维预测预警及应急响应机制研究

In order to properly address the issue of accurately predicting, efficiently early warning and reasonable risk response for aviation safety, time-dependent multivariate predicting model, the strategy for alerting and defense are studied in this program. Considering the characteristics of the randomness, relevance, complexity, time-dependence, etc. of the aviation safety problem comprehensively，this proposal will be implemented by the following steps. Based on the multivariate Gaussian process regression, the multivariate output prediction model of the aviation safety by considering the time-dependence and the uncertainty will be proposed, to explore the intrinsic relation between the causal factors and the aviation accidents, and then，the correlation of the causal factors in the time dimension will be established and analyzed. In addition, for the time-dependence of the multivariate prediction index, the early warning judgment strategy of aviation safety based on the multi-mode first passage principal will be proposed，to achieve quantitative early warning by virtue of exploring the relationship between the output response of early warning and the accident risk probability in the time dimension. Global sensitivity indices which reflect the contribution of component functions to the output will be proposed, and then used for ranking the contribution of the causal factors to the statistical distribution of multivariate alerting index. An optimization model which distribute the expectation performance of aviation safety to the control parameters of causal factors based on sensitivity and expectation distribution techniques will be proposed, then the distribution of causal factors variables is optimized gradually by the principle of global sensitivity function proportionality until aviation safety meet the design requirements, to achieve the inverse apportion for the safety expectation to the causal factors to obtain the quantitative and accurate defense. Take an aviation unit as an example, the theory and methodology are applied to test their feasibility and validity, and this program will provide theoretical basis and design methods for the prediction, early warning and defense of aviation safety, and is of great significance for completing the theoretical system of early warning and defense of aviation safety.

为提高航空安全预警预防水平应对航空业发展迅速事故隐患增大的现实问题，本研究考虑航空事故随机性、低频性、时变性特点，以事故预测预警及应急响应机制为研究对象，开展以下研究：（1）基于多维高斯过程建立考虑时变、不确定性因素的航空安全多维预测模型，揭示致因因素与事故之间的内在影响规律及其在时间维度的相关关系。针对多维预警指标时变规律，提出多模式首次超越预警策略，建立预警指标与事故风险在时间维度的映射关系，实现定量预警；（2）研究衡量致因因素对航空安全水平贡献程度的全局灵敏度指标及求解方法，基于致因因素对航空安全的贡献分解和精度分配研究安全风险向致因因素的逆向传递机理，构建安全期望向致因因素的逆向分配优化模型，完成安全期望向致因因素的逆向分配，实现精准响应；（3）对某航空单位开展安全预测预警及应急响应分析，推进工程应用。本项目为航空安全预警、预防提供理论方法，对完善航空安全管理理论体系具有重要意义。

本项目针对航空业发展迅速事故隐患增大的现实问题，研究了基于不确定性环境的航空预测预警方法及应急响应策略，提升航空安全科学管理水平。考虑考虑航空事故随机性、低频性、时变性特点以及众多致因因素的不确定性信息，开展了不确定性环境下航空安全信息识别，分析，并对其不确定性进行描述，研究了航空安全复杂系统的不确定传播分析，并对航空安全输出概率特性进行研究。在此基础上，基于机器学习、深度学习理论建立考虑时变、不确定性因素的航空安全多维预测模型，并对方法的效率和精度进行了交叉对比，揭示致因因素与事故之间的内在影响规律及其在时间维度的相关关系；针对不确定性对航空安全预测精度的影响，提出了考虑误差不确定性开展航空安全预测，通过非参数方法获得航空安全预测误差不确定性描述，基于最高密度域求解一定可靠程度航空安全预测值最可能落入区间，量化不确定性引起航空安全预测结果的变动，从而确定该区间包含航空安全预测值的可靠程度，更好地认识被预测量在未来变化中可能存在的不确定性和面临的风险；此外，考虑航空安全致因因素多，影响关系复杂，研究了基于互信息和灵敏度的航空安全预测精度贡献分析方法，衡量致因因素对航空安全的贡献大小，快速辨识对系统安全影响的关键影响因素，并探索了基于不确定性的航空安全预防对策生成机制。该项目所提方法、模型正确性和准确性在某航空单位开展了验证。并将该项目成果应用到2019和2020年度空军保障形势分析与工程测算任务，呈阅报告，受到空军首长肯定。通过项目研究，能够为航空事故预防，航空安全科学管理提供理论支撑和技术手段。