图模型与学习算法结合的贝叶斯网络自动建模

针对纯数据驱动的贝叶斯网络结构学习算法的准确度和效率较低的问题,提出了一种融合多信号流图模型与K2学习算法的贝叶斯网络自动建模方法。该方法利用多信号流图模型能够描述信号之间传递与依赖关系的能力,结合K2学习算法在结构学习中的优势,实现了专家知识与数据驱动方法有效融合的贝叶斯网络结构自动学习算法。通过与常用网络结构学习算法的对比实验证明,该融合算法显著降低了结构学习对学习范围和训练数据规模的要求,具有更高的学习准确度和运算效率。采用真实系统实例阐述了该融合算法的应用过程,验证了算法的实用性。      

考虑在轨运动可靠性的空间机械臂关节力矩优化方法

提出一种考虑在轨运动可靠性的冗余度空间机械臂关节力矩优化方法。首先将机械臂操作空间中点到点的转移任务从笛卡尔空间转换至关节空间,利用七次多项式插值法对各关节变量进行参数化处理,获得粒子群算法的优化控制参数;与传统路径规划方法不同,将机械臂各关节力矩的均值和最小作为粒子群算法优化求解的目标函数;依据该目标函数以及相应优化控制参数,利用粒子群算法对空间机械臂运行轨迹进行优化求解,得到机械臂关节力矩均值和最小的运行路径。仿真实验表明,相比传统路径规划方法及以关节力矩二范数为目标函数的关节力矩优化方法,在降低冗余度空间机械臂关节力矩均值方面,文中方法分别减小了33.57%和10.47%;在降低关节力矩最大值方面,分别减小了43.25%和6.19%。     

星载接收机的EMC分析与设计改进

文章介绍了星载接收机的EMC试验问题,分析问题产生的机理,提出星载接收机的EMC设计改进措施,并经过试验验证,改进措施切实有效.     

高速飞行器翼面结构热振动试验的TARMA模型方法

高速飞行器翼面结构的热振动试验研究对这类飞行器的设计和安全飞行具有重要的意义。采用时变自回归滑动平均(TARMA)模型方法建立了受热时变结构系统模态频率辨识的数学模型,并用一个数值算例进行了验证。将地面振动测试系统与瞬态热环境模拟系统相结合,设计了翼面结构热振动试验系统并模拟结构的瞬态温度场,同时对纯随机激振力激励下受热时变结构系统的振动位移信号进行测量,并用TARMA模型对时变固有频率进行了辨识,获得了前4阶固有频率随加热时间的变化规律,并将辨识结果与数值计算结果进行了比较,两者误差在5%以内。另外,在稳态均匀热环境下辨识得到的结构系统固有频率变化与数值计算结果也吻合得很好。通过将均匀温度场与瞬态温度场下的结果进行对比分析,指出了瞬态热环境下时变结构的固有频率随加热时间变化的趋势主要由结构材料属性的退化和结构内部不均匀热应力的影响共同决定。     

同心分级旋流结构的动态响应特征

实验探索了LESS(low emissions with stirred swirls)燃烧室头部出口附近的固有流动特征,及其对外界激励的响应.实验结果表明:预燃级旋流区域存在着周期性流动结构,而主燃级出口区域没有周期性流动结构;当外界激励的频率与流动固有频率耦合时,会引起预燃级流场畸变,并导致周期性流动的流速振荡幅值急剧增大.      

O2 evolution and cyclic electron flow around photosystem I in long-term ground batch culture of Euglena gracilis

Based on the purpose of better exploring the function of green producers in the closed aquatic biological life support system, the condition of dynamic O₂ evolution and performance of cyclic electron flow around photosystem I (CEF-PSI) in long-term ground batch culture of Euglena gracilis were studied, the relationship between linear electron flow (LEF) and CEF-PSI was revealed, the function of CEF-PSI was investigated. Excellent consistency in O₂ evolution pattern was observed in cultures grown in both closed and open containers, O₂ evolution was strictly suppressed in phase 1, but the rate of it increased significantly in phase 2. CEF-PSI was proposed to be active during the whole course of cultivation, even in the declining phase 3, it still operated at the extent of 47–55%. It is suggested that the relationship between LEF and CEF-PSI is not only competition but also reciprocity. CEF-PSI was proposed to contribute to the considerable growth in phase 1; it was also suggested to play an important protective role against photosystem II (PSII) photoinhibition at the greatly enhanced level (approximately 80–95%) on the 2nd day. Our results in this research suggest that E. gracilis had very particular photosynthetic characteristics, the strict O₂ evolution suppression in the initial culture phase might be a special light acclimation behavior, and CEF-PSI could be an important mechanism involved in this kind of adaptation to the changeable light environment.     

相位差法测量磁传感器阵列的非平行度

随着磁传感器技术及磁测技术的发展,通过测量磁梯度张量进行磁探测及磁定位已成为磁探领域的新方法。但在磁梯度张量测量过程中,由于机械安装的误差导致的磁传感器阵列中各传感器的磁轴的不平行将会极大的降低磁定位的精度。本文提出了一种易于实现的磁传感器阵列非平行度测量方法--相位差法,并通过试验验证了其有效性。     

贝叶斯网络结合决策理论的向前多步排故策略

针对序贯诊断、维修问题,提出基于贝叶斯网络和决策理论的向前多步排故策略生成算法.通过建立贝叶斯网络排故模型实现了不确定条件下排故知识的高效表达,同时使得推理算法与具体应用无关.采用决策影响图进行排故决策分析,充分利用观测操作间的相关性,选择合理的向前多步观测操作来降低维修盲目性.为了验证所提算法的有效性,采用随机排故策略、决策理论排故策略和理想排故策略的结果进行对比分析.仿真结果表明,所提算法通过增加合理观测操作,减少维修焦点和实际维修操作,使得总排故费用明显低于已有的启发式排故策略.      

故障预测算法稳定性实时评估方法

针对现有故障预测算法性能评估指标受实际剩余使用寿命约束的问题,从稳定性角度提出一种评估算法性能的方法。通过研究对象系统健康退化过程,在对象系统实际剩余使用寿命未知情况下,利用可以实时获得的剩余使用寿命预测值和已消耗寿命值,通过计算虚构寿命值的变异系数指标来客观评估故障预测算法的性能。为了验证所提方法的有效性,结合机电作动器故障演化模型仿真生成数据对递归最小二乘和粒子滤波两种故障预测算法的稳定性进行了实时评价。仿真结果表明,所提方法与运用已有指标、在获知剩余使用寿命理想值前提下得出的评估结果保持一致。     

Vehicle State and Parameter Estimation Based on Dual Unscented Particle Filter Algorithm

Acquisition of real-time and accurate vehicle state and parameter information is critical to the research of vehicle dynamic control system. By studying the defects of the former Kalman filter based estimation method, a new estimating method is proposed. First the nonlinear vehicle dynamics system, containing inaccurate model pa rameters and constant noise, is established. Then a dual unscented particle filter （DUPF） algorithm is proposed. In the algorithm two unscented particle filters run in parallel, states estimation and parameters estimation update each other. The results of simulation and vehicle ground testing indicate that the DUPF algorithm has higher state estimation accuracy than unscented Kalman filter （UKF） and dual extended Kalman filter （DEKF）, and it also has good capability to revise model parameters.