基于多时间尺度特征的语音识别模型

提出了基于多时间尺度特征的语音识别模型.该模型采用描述谱参数轨迹的段特征,在段尺度上实现了对语音信号帧间相关性的显式建模;采用段特征依赖的非平稳时间序列产生模型,实现了不同尺度特征间的相关性建模,并在帧尺度上通过参数化的均值轨迹函数,实现了对语音信号帧间相关性的隐式建模.给出了基于多时间尺度特征联合统计距离优化的分段算法及基于最大似然准则的模型参数估计算法.识别实验表明,该模型的识别性能优于标准HMM及趋势HMM.     

确定时间序列协方差函数的方法

提出一种确定时间序列协方差函数的方法,它首先根据(多元)时间序列构造其互协方差函数随机序列、互相关函数随机序列或自协方差函数随机序列、自相关函数随机序列,然后采用谱分析和多点平均方法对互协方差函数随机序列、互相关函数随机序列或自协方差函数随机序列、自相关函数随机序列的趋势项进行分离,分别求得其周期项和非周期项的函数表达式,再综合给出整个趋势项函数。从而得到原时间序列的互协方差函数、互相关函数或自协方差函数、自相关函数的函数形式,并通过最小二乘方法确定其中的待定参数。该方法可用于时间序列协方差函数的建模、分析和预测,并且计算简单易行、精度高,便于实际应用。      

Gevers—Wouters算法在动态数据建模中的应用研究

现代时间序列分析方法是解决最优滤波问题的不同于Winner滤波和Kal man滤波的方法,其基本分析方法是AR MA新息模型。而构造ARMA新息模型可归结为求MA参数。本文主要研究了直接运用Gevers-Wouters算法确定观测信号模型参数的方法,使得在进行数据建模和构造ARMA新息模型时,可统一运用Gevers-Wouters算法。运用此方法对校准某型机载雷达的实测数据进行处理,结果显示出快速收敛性。     

钉载作用下层板多孔过盈配合分析

本文利用各向异性体平面弹性理论中的复势方法,以Faber级数为工具,导出层板多孔/弹性核过盈配合在钉载作用下的级数解。     

Black-hole transients: From QPOs to relativistic jets

Due to the impressive amount of new data provided by the RXTE satellite in the past decade, our knowledge of the phenomenology of accretion onto black holes has increased considerably. In particular, it has been possible to schematize the outburst evolution of transient systems on the basis of their spectral and timing properties, and link them to the ejection of relativistic jets as observed in the radio. Here, I present this scheme, concentrating on the properties of the quasi-periodic oscillations observed in the light curves and on the link with jet ejection.     

SINS基于非线性量测的大失准角初始对准算法

针对陆用捷联惯导系统(SINS)大失准角初始对准问题,提出一种基于二阶非线性量测的二阶泰勒级数展开的初始对准算法。首先,将大失准角非线性对准问题,转化为具有线性系统方程和二阶非线性量测方程的滤波问题。然后,设计基于二阶非线性量测方程的二阶泰勒级数展开的滤波算法,可以在任意姿态和无初值条件下进行初始对准和参数估计。最后进行车载对准试验,结果表明,新算法可以在300s内完成实现大失准角初始对准,对准精度与传统分段对准算法相当。     

确定时间序列均值和方差函数的移动多点平均方法

为了提高有限样本或小样本情况下时间序列均值和方差函数的确定精度,以保证时序建模、分析和预测精度,提出一种确定序列趋势项的移动多点平均方法.该方法能够得到时间序列均值中非周期部分,结合样本周期图法得到的周期项,可综合得到其均值函数,并可进一步得到时序的标准差函数.Monte Carlo模拟对比结果表明:该方法不仅能够保证时间序列段内分析精度,而且能够有效提高时间序列的预报精度.      

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。     

Monitoring of the space system

We have performed monitoring of the space system that includes the Earth, the Moon, the Sun, and the GPS satellite group. We have discovered semi-diurnal and diurnal periodicities in the number of satellites detected as well as in the altitude, latitude and longitude by a GPS receiver. We have revealed tidal deformations related to changes in the Earth?s orientation with respect to the Moon and the Sun.     

DPOD2014: A new DORIS extension of ITRF2014 for precise orbit determination

To support precise orbit determination of the altimetry missions, the International DORIS Service (IDS) regularly estimates the DPOD (DORIS terrestrial reference frame for Precise Orbit Determination) solution which includes mean positions and velocities of all the DORIS stations. This solution is aligned to the current realization of the International Terrestrial Reference Frame (ITRF) and so, can be seen as a DORIS extension of the ITRF. In 2016, moving to the IDS Combination Center, the DPOD construction scheme changed. The new DPOD solution is produced from a DORIS cumulative position and velocity solution. We present the new methodology used to compute DPOD2014 and its validation procedure. In order to present geophysical applications and interpretations of these results, we show two examples: (1) the Gorkha earthquake (M7.8 – April 2015) generates a 3-D mis-positioning of nearly 55?mm of the EVEB DORIS station at the Everest base camp 90?km from the epicenter. (2) Applying the results the DPOD2014 realization, we show that the most recent vertical velocity of Thule, Greenland is similar to that observed between 2006 and 2010, indicating further ongoing ice mass loss in the Thule region of northwest Greenland.