一种非规则采样航空时序数据异常检测方法

航天器遥测数据的实时异常检测对于航天任务具有重要意义。以往方法大都考虑规则采样且缺失率较低的时序数据,然而航空时序数据具有维度大、噪声多、缺失率高、采样间隔不规则等特点,因此异常检测任务较为困难。针对非规则采样且具有缺失值的多维航空时序数据提出非规则采样多维时序数据异常检测（IMAD）算法。首先,采用带有可训练迟滞项的门控循环单元（GRU-D）对缺失值和非规则采样的时序数据进行建模;然后,采用变分自编码器建立随机性模型,学习正常时序数据的分布,从而对噪声数据具有鲁棒性;最后,利用基于极值理论的自适应阈值确定法确定合适阈值进行异常检测。结果显示,在两个真实航空时序数据集上,IMAD具有超出当前最新异常检测算法的性能;多个实验表明,IMAD在缺失率、参数以及数据集变化时,能够维持较好的异常检测效果,具有较强的鲁棒性。     

Equatorial Ionospheric Anomaly (EIA) and comparison with IRI model during descending phase of solar activity (2005–2009)

The ionospheric variability at equatorial and low latitude region is known to be extreme as compared to mid latitude region. In this study the ionospheric total electron content (TEC), is derived by analyzing dual frequency Global Positioning System (GPS) data recorded at two stations separated by 325 km near the Indian equatorial anomaly region, Varanasi (Geog latitude 25°, 16^/ N, longitude 82°, 59^/ E, Geomagnetic latitude 16°, 08^/ N) and Kanpur (Geog latitude 26°, 18^/ N, longitude 80°, 12^/ E, Geomagnetic latitude 17°, 18^/ N). Specifically, we studied monthly, seasonal and annual variations as well as solar and geomagnetic effects on the equatorial ionospheric anomaly (EIA) during the descending phase of solar activity from 2005 to 2009. It is found that the maximum TEC (EIA) near equatorial anomaly crest yield their maximum values during the equinox months and their minimum values during the summer. Using monthly averaged peak magnitude of TEC, a clear semi-annual variation is seen with two maxima occurring in both spring and autumn. Results also showed the presence of winter anomaly or seasonal anomaly in the EIA crest throughout the period 2005–2009 only except during the deep solar minimum year 2007–2008. The correlation analysis indicate that the variation of EIA crest is more affected by solar activity compared to geomagnetic activity with maximum dependence on the solar EUV flux, which is attributed to direct link of EUV flux on the formation of ionosphere and main agent of the ionization. The statistical mean occurrence of EIA crest in TEC during the year from 2005 to 2009 is found to around 12:54 LT hour and at 21.12° N geographic latitude. The crest of EIA shifts towards lower latitudes and the rate of shift of the crest latitude during this period is found to be 0.87° N/per year. The comparison between IRI models with observation during this period has been made and comparison is poor with increasing solar activity with maximum difference during the year 2005.     

Determination of the mean dynamic ocean topography model through combining multi-source gravity data and DTU15 MSS around China's coast

Mean dynamic ocean topography (or MDT) is closely related to ocean circulation and global climate change. It has important scientific significance and application value for the development and utilization of marine resources in China's coastal areas. Based on the terrain gravity, marine gravity, and SRTM 3?s data, an algorithm to reduce the problem of gravity data gaps between land and sea is proposed. A consistent land-sea gravity model is established based on point-mass fusion method. Then geoid model, which accuracy was estimated to be 8.5?cm through the verification of 348 GNSS/level data from the coastal provinces, of China's coastal areas was calculated through remove-restore technique. Connecting the above geoid model with DTU15 MSS model to establish a MDT model in China's coastal areas using the direct method in space domain. The effect of gravity field model, dominant factors of sea surface topography, and low pass filter are analyzed. Taking Bohai Sea and Yellow Sea as an example, and comparing MDT with the two international models CNES_2013_MDT and DTU15_MDT. The results show that the MDT has the potential to construct a vertical datum of the ocean and carry out related scientific research and application.     

基于传感器关联网络的燃气轮机异常检测方法

通过分析燃气轮机4种典型异常形式的产生机理以及特征表现,得到了不同异常形式与传感器关联网络结构特征之间的映射关系。在此基础上,得到传感器关联网络的异常特征模式,提出了基于传感器关联网络的燃气轮机异常检测策略。通过实例分析,证明了融合多源信息基础上建立的传感器关联网络模型,可以过滤掉节点间关联指标低于阈值037的相关性,有效的实现燃气轮机的稳态异常检测,并可以检测出转速上升过程中,节点间相关性的正常线性变化趋势,存在大于12%的异常凹陷非线性趋势,从而有效的实现燃气轮机的动态异常检测。      

姿态异常下的星体自旋角速度确定方法

摘要： 针对在轨姿态异常后星体高速自旋情况,开展星体自旋角速度确定方法研究.提出一种基于太阳敏感器测量数据确定自旋角速度的方法,并对自旋角速率确定误差进行分析,得到各误差源对确定精度的影响关系;基于角速率确定误差分析结论,给出根据角速度大小选取不同时间间隔测量数据的策略,通过大时间间隔测量数据的选取,保证小角速度情况下的确定精度.所提出方法的有效性通过数学仿真验证,并在实际应用中基于太阳敏感器遥测数据获取在轨姿态异常卫星角速度.     

机载武器作战使用可靠性综合验证

为减少机载武器可靠性验证试验样本量和降低验证试验风险,通过对机载武器作战使用可靠性验证的工程需求、试验方案、统计模型、验证方法等进行研究,提出了一种利用研制试验数据和定型飞行试验数据的作战使用可靠性综合验证方法。针对机载武器的特点,根据变动统计理论,考虑性能验证与可靠性验证的要求,综合利用研制试验数据和定型飞行试验数据,给出作战使用可靠性综合验证方案,进而制定设计定型飞行试验方案。实例表明,在不改变生产方风险和使用方风险的情况下,该方法可有效地验证机载武器挂飞可靠性和自主飞行可靠性指标,显著地减少定型飞行试验样本量,大幅节约研制成本,并缩短研制周期。     

Tool Materials Rapid Selection Based on Initial Wear

Based on a large amount of literature about tool wear research, 873 tool wear curves are taken as samples, and statistical analysis is carried out to select the most suitable tool from all the tool materials suggested by the tool manufacturers. Statistical relationships between the initial wear and uniform wear periods are obtained. The results show that there is qualitative relationship between wear rate during initial wear period (WRIWP) and wear rate in uniform wear period (WRUWP) to certain extent. On this basis, a tool material rapid selection method based on the initial wear is put forward, and suitable tool materials for machining titanium alloy are selected. The experimental results indicate that this method is effective and useful. The new tool materials rapid selection can be used to select suitable cutting tool materials quickly before carrying out systematic machinability tests with the most suitable tool materials. The technology can be applied to doing the initial selection of cutting tool materials in either the machinability research or the workshop production.     

低高度内辐射带高能质子空间分布位形的研究

利用国际地磁参考场模式(IGRF模式)分析了1970-2000年低高度南大西洋负磁异常区位形的漂移与变化，给出了几个高度异常区中心位置磁场强度的变化和位置的变化。利用带电粒子的运动学方程，简要分析了低高度辐射带高能粒子的运动，得出在低高度，磁场是决定辐射带高能粒子空间强度与分布的决定性因素．低高度辐射带空间分布位形的变化特征应该与低高度南大西洋负磁异常区的变化特征基本一致．低高度南大西洋负磁异常区的特征可以作为低高度辐射带空间分布位形的一个初步判据。     

基于增量式孤立森林的液体火箭发动机异常检测方法

为解决液体火箭发动机故障标签缺失条件下流数据无监督检测问题,以及满足不同发动机台次和不同工况的自适应检测需求,基于增量学习思想,提出了基于增量式孤立森林的异常检测算法。设计了多工况流数据检测条件下的在线更新策略、异常分数表达式,并通过更新停止策略避免故障数据对模型的污染。利用多台次试车数据对该模型进行验证,并与传统方法进行比较,结果表明,该算法能够对样本异常程度进行量化评价,能够有效检测早期缓变故障,其F₁指标较原始孤立森林算法提高了43%,检测及时性优于红线算法和自适应阈值算法。