含未知输入非线性系统的扩展平方根容积卡尔曼滤波算法

针对工程实际应用中存在的未知输入会导致经典的非线性滤波器状态估计精度下降甚至滤波发散的问题,提出了一种基于最小方差无偏估计(minimum variance unbiased estimation, MVUE)准则的扩展平方根容积卡尔曼滤波(extended square-root cubature Kalman filter, ESRCKF)算法。首先,结合上一时刻未知输入估计值对状态一步预测值进行修正,得到含未知输入条件下的状态预测值。其次,设计新息并采用加权最小二乘(weighted least squares, WLS)法获取当前时刻未知输入的无偏估计。最后,通过最小化协方差矩阵的迹,同时采用拉格朗日乘子法和舒尔补引理得到系统状态的最小方差无偏估计。仿真结果表明,相比于现有的非线性滤波算法,ESRCKF算法提高了在处理含未知输入非线性系统时的状态估计精度,并能同时实现系统状态和未知输入的最优估计,验证了该算法的有效性。     

离子推力器栅极组件热变形测量方法

栅极组件是离子推力器的核心部件,其所包含的多层栅极结构在工作过程中会发生热变形,导致栅极间距发生改变,进而影响离子推力器的工作性能及寿命。为了实现栅极结构在实际工况下热变形的非接触测量与评估,本研究提出了基于摄像测量的栅极变形观测方法,评价了不同图像分析方法（灰度重心法、圆拟合法、灰度匹配法）对变形测量精度的影响。结果表明灰度匹配方法观测精度高,环境适应性好;同时,设计并实现了一套栅极热变形环境模拟与变形测量系统,可实现高温、通水、通气环境下变形的高精度观测。最后,对环境箱通水通气对测量精度的影响进行了评价,结果表明二者均会对测量精度造成影响,通水对实验结果的影响大于通气的影响。     

基于时延嵌入式隐马尔科夫模型的癫痫脑电分类算法

癫痫脑电的分类识别能够为癫痫的预警和病程的发展监测提供强有力的技术支持。传统的癫痫脑电分类识别方法需要从较长的时间序列中提取特征,难以刻画大脑的瞬态变化,检测低效且耗时,降低了癫痫预警的有效性。针对上述问题,提出一种基于隐马尔科夫模型的癫痫脑电分类算法。该方法通过时延嵌入式隐马尔科夫模型(time-delay embedded hidden Markov model,TDE-HMM)对脑电进行状态估计,并提取状态序列中的状态切换特征,通过多层感知机(multiple layer perceptron,MLP)实现对不同癫痫发作阶段脑电的有效辨识。实验结果表明,相较于小波变换、微分熵等传统特征,所提方法准确率高,能够有效刻画癫痫不同阶段的大脑状态变化,为癫痫脑电的分类识别和状态分析提供了新的备选方案。     

有色噪声中多个周期相近信号参数的估计算法

本文提出了一种“谐波组清除算法”来估计周期信号的参数。该算法不仅计算简单并可获得较高的频率分辨率,而且对信号模型具有较强的鲁棒性。最后,用数值仿真数据与实测直升机噪声数据对该算法进行了检验。     

连续竞价市场中格式化特征形成机理的仿真研究

现有的研究一般将金融市场中的胖尾分布、波动聚集、长期记忆等格式化特征的形成机理归于交易者的行为.为了研究市场交易机制对这些特征有何影响,以我国证券市场交易制度为蓝本,建立了一个连续竞价市场的基于Agent的仿真模型.仿真结果表明:在连续竞价机制下,即使Agent是理性的基本分析者或简单的随机交易者,模型仍然能够再现上述特征.这一结果说明,交易机制也是导致格式化特征的重要原因.     

Novel cued search strategy based on information gain for phased array radar

A search strategy based on the maximal information gain principle is presented for the cued search of phased array radars. First, the method for the determination of the cued search region, arrangement of beam positions, and the calculation of the prior probability distribution of each beam position is discussed. And then, two search algorithms based on information gain are proposed using Shannon entropy and Kullback-Leibler entropy, respectively. With the proposed strategy, the information gain of each beam position is predicted before the radar detection, and the observation is made in the beam position with the maximal information gain. Compared with the conventional method of sequential search and confirm search, simulation results show that the proposed search strategy can distinctly improve the search performance and save radar time resources with the same given detection probability.     

Cluster Synchronization in a Complex Dynamical Network with Two Nonidentical Clusters

This paper further investigates cluster synchronization in a complex dynamical network with two-cluster. Each cluster contains a number of identical dynamical systems, however, the subsystems composing the two clusters can be different, i.e., the individual dynamical system in one cluster can differ from that in the other cluster. Complete synchronization within each cluster is possible only if each node from one cluster receives the same input from nodes in other cluster. In this case, the stability condition of one-cluster synchronization is known to contain two terms： the first accounts for the contribution of the inner-cluster coupling structure while the second is simply an extra linear term, which can be deduced by the ＂same-input＂ condition. Applying the connection graph stability method, the authors obtain an upper bound of input strength for one cluster if the first account is known, by which the synchronizability of cluster can be scaled. For different clusters, there are different upper bound of input strength by virtue of different dynamics and the corresponding cluster structure. Moreover, two illustrative examples are presented and the numerical simulations coincide with the theoretical analysis.     

要加强对复杂适应性系统的研究——对自组织理论的反思

简单回顾了自组织理论的发展历史、内容和采用的方法,重点分析了自组织理论的研究对象和研究范围;指出自组织理论在研究复杂系统上的局限性——只适于研究简单巨系统,它研究的是系统演化的结果而非过程;自组织理论使用微分方程作为数学工具,要求子系统之间的相互作用要满足叠加原理而不考虑涌现的情况。在此基础上,本文提出要加强复杂适应性系统的研究,具体有两方面的工作：首先要加强计算机作为数学工具的应用,扩展对系统状态描述的范围;另外要加强对层次性的研究,重点是对涌现问题的分析。最后提出现在加强复杂系统的研究应从实际的具体系统入手。     

红外探测器成像仿真

红外探测器是红外成像技术的核心,直接影响着红外成像系统的性能。在分析目标成像尺寸的基础上,详细分析光学弥散、探测器噪声、成像阵列盲元等的形成机理,并进行建模和仿真,最后将这些模型依次叠加。采用此方法得到的仿真图像充分考虑了探测器各种效应的影响,提高了仿真图像的真实性与逼真度,可为红外成像系统的研制及目标检测识别算法提供测试场景与依据。

Abstract：

Infrared detector, which can directly influence the performance of infrared imaging system, plays a key role in infrared imaging technology. The scale of the target image, the causation of optics dispersion, noise of the detector and blind pixels in the imaging arrays were detailed analyzed. More actual infrared images would be simulated by adding the effects mentioned above to the theoretical image model, because the ill effects were calculated. The simulated images could be used to test infrared imaging system or the algorithm of target detecting and recognition.     

组合聚类与最优核v-SVR融合方法在能源系统应用

针对能源复杂系统样本数量有限、变量维数高、偶合关系复杂等问题,提出了一种组合聚类算法和最优核v-支持向量回归机SVR融合的方法。该方法采用SOM自组织映射神经网络和K-means组合的聚类算法对初始样本集合进行聚类,构成不同核函数的子支持向量回归机SVR模型,再用均方误差标准（MSE）和小误差概率对其各核函数进行优选,得到最优核函数的v-支持向量回归机SVR模型。仿真结果表明,采用这种方法进行能源供需预测是有效的,其精度优于常规的支持向量回归建模方法。

Abstract：

A method of fusion strategy with Optimal Kernel v-SVR （Support Vector Regression） and assembled clustering algorithm was proposed,facing the issues of complex energy system like limited samples,high dimension,complex coupling.The assembled clustering algorithm was used to cluster the initial samples related to energy data to form sub-SVR models based on different Kernel functions,combining Self-Organizing Map （SOM） neural network with K-means algorithm.The Mean Square Error criterion （MSE） and small error probability were used to evaluate the Kernel functions to obtain the optimal Kernel v-SVR model.The simulation results demonstrate that the proposed method is valid for predicting energy supply/demand and its accuracy is superior to the conventional SVR method.