多源不确定性下基于海洋观测序列的定点观测设备故障预测研究

This proposal is motivated by the development of global ocean observing network. This work is designed to tackle the problems existing in the maintenance of observing equipment in complex marine systems by relying on China's National Shallow Sea Testing Field in Weihai, Shandong Province. With the mooring floats for offshore fixed-point observation as the research object, this study emphasizes the accuracy, reliability and timeliness of observing equipment fault prediction by using observing sequences of marine environmental factors under multisource uncertainties. It also focuses on the establishment of a natural background interval with observing sequences of marine environmental factors, the extraction of characteristics in background response relationships with a multiple coordinated observing sequences, as well as the expression of the response relationships between observing equipment components and anomalies in environmental observation. This program will highlight (1) the integrated research method covering investigation and consultation, systematic analysis, numerical simulation, and example verification, so as to pursue an approach for effectively identifying, quantifying and integrating multisource uncertainties in the process of fault prediction for ocean observing equipment; (2) the construction method to reflect the natural background state of the principles, structures and inherent laws of marine environmental change; as well as (3) the evaluating and analyzing method for fault prediction results using observing sequences of marine environmental factors. This study is conducive to optimizing the strategies for regular maintenance of ocean observing equipment and avoiding repeated repairs after interruption due to faults, with a view to reducing the costs and losses in ocean observation and enhancing the safety, reliability and economic efficiency in the operation of China's ocean observing system.

基于全球海洋观测网建设背景，针对海洋复杂系统下观测设备运行保障存在的问题，依托于我国国家浅海海上试验场条件，以离岸定点观测的锚系浮标为研究对象，注重在多源不确定性下利用海洋环境要素观测序列进行观测设备故障预测的准确性、可靠性和及时性，聚焦海洋环境要素观测序列自然背景区间建立、多协同观测要素序列自然背景响应关系特征提取、观测设备部件同环境要素观测异常之间响应关系表达，着眼调研咨询、系统分析、数值模拟、实例验证相结合的研究方法，致力研究海洋观测设备故障预测过程中多源不确定性的有效识别、量化和融合方法，反映海洋环境变化规则、结构和内在规律自然背景状态的构建方法，以及利用海洋环境要素观测序列对观测设备故障预测结果的评价分析方法。利于改变目前我国海洋观测装备计划性定期维护或故障导致观测中断后再维修的策略，降低海上观测成本、减少损失，提高我国海洋观测系统业务运行的安全性、可靠性和经济性。

针对海洋复杂系统下定点观测设备运行保障存在的问题，本项目获取了威海国家海洋综合试验场背景场观测系统中锚系浮标观测的水文气象关键环境要素的时间序列和真实故障状态序列；设计构建了基于海洋观测序列的海洋定点观测设备故障识别框架和技术方法，分析了自然异常序列和设备故障异常序列对不同模型结构、检测策略以及分类水平等因素的敏感性，探讨了模型精度、不确定性以及先验概率等对观测序列异常检测和设备故障诊断结果的影响。所得出主要结论（成果）包括：（1）耦合观测值（OS）和预测值（PS）窗口检测策略的小波神经网络海洋定点观测序列异常检测方法，可模拟异常数据较稀缺的海洋环境系统的正常行为，不需标记训练数据、可在无监督情况下工作；选择99%的背景区间阈值划分水平可保证较高判定可靠性；利用OS策略和正态分布划定残差阈值可有效检测自然因素引起的偶发性、回复性异常，而采用PS策略和更靠近残差均值中心的逻辑分布判定可以有效检测设备因素导致的持续性、不可逆异常。（2）可通过多要素序列的弹性关联判定强化单要素序列异常检测的结果；耦合改进的符号近似距离和动态时间规整相关性度量算法有效解决了海洋自然环境变化导致的多要素时间序列间扭曲、伸缩和异位，明确了选择6-8的分段子长度值和10个滑动窗口宽度对检测效果和成本是合理的折中，能够达到工程应用。（3）设计构建的锚系浮标贝叶斯网络故障诊断模型可直接利用海洋观测序列异常检测信息判定各部件相关故障的发生概率，不需额外设备运行状态数据，不确定信息下适当降低泄漏概率至0.5%可保障故障诊断结果的可靠性。（4）项目获得的国家发明专利针对业务化海洋观测网中的定点观测系统具有较好的普适性和工程应用准确性，当前部分算法正在国家海洋观测网设备状态运行监控系统中调试应用。