基于分布式神经网络递推预报误差算法的非线性系统建模

采用基于递推预报误差算法的分布式神经网络 结构建立非线性系统模型．子神经网络模型及其连接权值均采用递推预报误差方法来进行训 练,将所有子网络融合得到的分布式神经网络模型在模型精确性和鲁棒性方面有显著地增加 ．该方法较好地应用于复杂非线性动态系统的建模．     

Nonlinear nonnegative matrix factorization based on Mercer kernel construction

Generalizations ofnonnegative matrix factorization (NMF) in kernel feature space, such as projected gradient kernel NMF (PGKNMF) and polynomial Kernel NMF (PNMF), have been developed for face and facial expression recognition recently. However, these existing kernel NMF approaches cannot guarantee the nonnegativity of bases in kernel feature space and thus are essentially semi-NMF methods. In this paper, we show that nonlinear semi-NMF cannot extract the localized components which offer important information in object recognition. Therefore, nonlinear NMF rather than semi-NMF is needed to be developed for extracting localized component as well as learning the nonlinear structure. In order to address the nonlinear problem of NMF and the semi-nonnegative problem of the existing kernel NMF methods, we develop the nonlinear NMF based on a self-constructed Mercer kernel which preserves the nonnegative constraints on both bases and coefficients in kernel feature space. Experimental results in face and expressing recognition show that the proposed approach outperforms the existing state-of-the-art kernel methods, such as KPCA, GDA, PNMF and PGKNMF.     

Dynamic neural networks with hybrid structures for nonlinear system identification

Dynamic neural networks (DNNs) have important properties that make them convenient to be used together with nonlinear control approaches based on state space models and differential geometry, such as feedback linearisation. However the mapping capability of DNNs are quite limited due to their fixed structure, that is, the number of layers and the number of hidden units. An example shown in this paper has demonstrated this limitation of DNNs. The development of novel DNN structures, which has good mapping capability, is a relevant challenge being addressed in this paper. Although the structure is changed minorly only, the mapping capability of the new designed DNN in this paper has been improved dramatically. Previous work [J. Deng et al., 2005. The dynamic neural network of a hybrid structure for nonlinear system identification. In: 16th IFAC World Congress, Prague.] presents a new dynamic neural network structure which is suitable for the identification of highly nonlinear systems, which needs the outputs from the real system for training and operation. This paper presents a hybrid dynamic neural network structure which presents a similar idea of serial–parallel hybrid structure, but it uses an output from another neural network for training and operation classified as a serial–parallel model. This type of DNNs does not require the output of the plant to be used as an input to the model. This neural network has the advantages of good mapping capabilities and flexibilities in training complicated systems, compared to the existed DNNs. A theoretical proof showing how this hybrid dynamic neural network can approximate finite trajectories of general nonlinear dynamic systems is given. To illustrate the capabilities of the new structure, neural networks are trained to identify a real nonlinear 3D crane system.     

基于Tesla平台的快速反投影运算

反投影运算是锥束CT图像重建算法中运算量最大,最耗时的部分,是制约重建速度的瓶颈。为此,在计算统一设备架构模型下,应用体素驱动法实现基于Tesla平台的反投影(BP)并行运算,并对BP运算上的访存和数学指令进行优化。实际CT数据的重建结果表明,该方法的运算速度是CPU串行程序的198倍,效率高且易于实现。     

Parallel implementation of multilayered neural networks based on Map-Reduce on cloud computing clusters

To meet the requirements of big data processing, this paper presents an efficient mapping scheme for a fully connected multilayered neural network, which is trained by using back-propagation (BP) algorithm based on Map-Reduce of cloud computing clusters. The batch-training (or epoch-training) regimes are used by effective segmentation of samples on the clusters, and are adopted in the separated training method, weight summary to achieve convergence by iterating. For a parallel BP algorithm on the clusters and a serial BP algorithm on an uniprocessor, the required time for implementing the algorithms is derived. The performance parameters, such as speedup, optimal number and minimum of data nodes are evaluated for the parallel BP algorithm on the clusters. Experiment results demonstrate that the proposed parallel BP algorithm in this paper has better speedup, faster convergence rate, less iterations than that of the existed algorithms.     

An efficient parameterization of dynamic neural networks for nonlinear system identification

Dynamic neural networks (DNNs), which are also known as recurrent neural networks, are often used for nonlinear system identification. The main contribution of this letter is the introduction of an efficient parameterization of a class of DNNs. Having to adjust less parameters simplifies the training problem and leads to more parsimonious models. The parameterization is based on approximation theory dealing with the ability of a class of DNNs to approximate finite trajectories of nonautonomous systems. The use of the proposed parameterization is illustrated through a numerical example, using data from a nonlinear model of a magnetic levitation system.     

The quadratic method for computing the eigenpairs of a martix

In this paper we introduce a new method for computing the eigenpairs of a matrix. The quadratic method which has the advantage of working in parallel, is based on solving quadratic nonlinear systems. The starting values of Newton's method used to solve the systems, are obtained by using a homotopy method together with polynomial interpolation. The algorithm is described and several numerical examples are given.     

Optimized OpenCL implementation of the Elastodynamic Finite Integration Technique for viscoelastic media

Development of parallel codes that are both scalable and portable for different processor architectures is a challenging task. To overcome this limitation we investigate the acceleration of the Elastodynamic Finite Integration Technique (EFIT) to model 2-D wave propagation in viscoelastic media by using modern parallel computing devices (PCDs), such as multi-core CPUs (central processing units) and GPUs (graphics processing units). For that purpose we choose the industry open standard Open Computing Language (OpenCL) and an open-source toolkit called PyOpenCL. The implementation is platform independent and can be used on AMD or NVIDIA GPUs as well as classical multi-core CPUs. The code is based on the Kelvin–Voigt mechanical model which has the gain of not requiring additional field variables. OpenCL performance can be in principle, improved once one can eliminate global memory access latency by using local memory. Our main contribution is the implementation of local memory and an analysis of performance of the local versus the global memory using eight different computing devices (including Kepler, one of the fastest and most efficient high performance computing technology) with various operating systems. The full implementation of the code is included.     

基于差分进化的BP神经网络预测混沌时间序列

针对BP神经网络预测模型收敛速度慢和容易陷入局部极小值的缺点,将差分进化算法和神经网络结合起来,提出了一种基于差分进化算法的BP神经网络预测混沌时间序列的方法,利用差分进化算法的全局寻优能力对BP神经网络的权值和阈值进行优化,然后训练BP神经网络预测模型求得最优解,将该预测方法用到3个典型的混沌时间序列进行算法的有效性验证,并与BP算法的预测精度进行了比较,仿真结果表明该方法对混沌时间序列预测具有更好的非线性拟合能力和更高的预测准确性。     

基于Hadoop平台的LDA算法的并行化实现

随着互联网的飞速发展,需要处理的数据量不断增加,在互联网数据挖掘领域中传统的单机文本聚类算法无法满足海量数据处理的要求,针对在单机情况下,传统LDA算法无法分析处理大规模语料集的问题,提出基于MapReduce计算框架,采用Gibbs抽样方法的并行化LDA主题模型的建立方法。利用分布式计算框架MapReduce研究了LDA主题模型的并行化实现,并且考察了该并行计算程序的计算性能。通过对Hadoop并行计算与单机计算进行实验对比,发现该方法在处理大规模语料时,能够较大地提升算法的运行速度,并且随着集群节点数的增加,在加速比方面也有较好的表现。基于Hadoop平台并行化地实现LDA算法具有可行性,解决了单机无法分析大规模语料集中潜藏主题信息的问题。     

A coarse-grained parallel approach for seismic damage simulations of urban areas based on refined models and GPU/CPU cooperative computing

Refined models and nonlinear time-history analysis have been important developments in the field of urban regional seismic damage simulation. However, the application of refined models has been limited because of their high computational cost if they are implemented on traditional central processing unit (CPU) platforms. In recent years, graphics processing unit (GPU) technology has been developed and applied rapidly because of its powerful parallel computing capability and low cost. Hence, a coarse-grained parallel approach for seismic damage simulations of urban areas based on refined models and GPU/CPU cooperative computing is proposed. The buildings are modeled using a multi-story concentrated-mass shear (MCS) model, and their seismic responses are simulated using nonlinear time-history analysis. The benchmark cases demonstrate the performance-to-price ratio of the proposed approach can be 39 times as great as that of a traditional CPU approach. Finally, a seismic damage simulation of a medium-sized urban area is implemented to demonstrate the capacity and advantages of the proposed method.     

基于网格的分布并行计算策略

网格计算为用户处理很多复杂问题提供了新方法，用网格实现大规模分布并行计算是必然的趋势。文章给出了基于网格中间件Globus Toolkit来实现分布并行计算的两种策略：紧密耦合并行程序和松散耦台并行服务，并给出实现这两种策略的实现方法，为实现分布并行计算提供了新方法。最后根据所提出的并行计算策略，在网格计算环境下实现了一个分布并行计算实例，并给出了相应的试验结果。     

基于PSO-DE-BP的光伏发电功率短期预测

提高光伏发电功率预测精度对保障智能电网安全稳定运行有重要意义。针对传统BP神经网络存在预测精度不高且收敛速度慢的弊端,提出一种基于粒子群（PSO）差分进化（DE）并行计算优化BP神经网络的光伏发电短期预测方法。首先分析影响因素重要程度,通过带权重的欧式距离筛选相似的训练样本集。其次,对粒子群分组,通过粒子群和差分进化混合算法对粒子组内和组间优化,以保证种群多样性、提高预测稳定和精度、避免局部最优。然后,建立预测模型,通过基于spark的内存计算平台,将PSO-DE-BP算法并行优化以提高算法运行效率。最后,根据不同天气类型的预测结果对模型进行分析验证,此方法比PSO-BP、BP算法模型具有更高的稳定性和预测精度。     

求解半线性椭圆型方程组的并行算法

本文结合区域分裂技术、多重网格方法、加速Ｓｃｈｗａｒｚ收敛方法、高低解方法、非线性Ｊａｃｏｂｉ迭代方法和Ｎｅｗｔｏｎ线性化迭代方法,设计了三种求解半线性椭圆型方程（组）的并行算法：并行Ｎｅｗｔｏｎ多重网格算法、并行非线性多重网格算法和并行加速Ｓｃｈｗａｒｚ收敛算法。数值试验说明这三种算法的并行计算是可行的。     

Using multiple levels of parallelism to enhance the performance of domain decomposition solvers

Large-scale scientific simulations are nowadays fully integrated in many scientific and industrial applications. Many of these simulations rely on modelisations based on PDEs that lead to the solution of huge linear or nonlinear systems of equations involving millions of unknowns. In that context, the use of large high performance computers in conjunction with advanced fully parallel and scalable numerical techniques is mandatory to efficiently tackle these problems.In this paper, we consider a parallel linear solver based on a domain decomposition approach. Its implementation naturally exploits two levels of parallelism, that offers the flexibility to combine the numerical and the parallel implementation scalabilities. The combination of the two levels of parallelism enables an optimal usage of the computing resource while preserving attractive numerical performance. Consequently, such a numerical technique appears as a promising candidate for intensive simulations on massively parallel platforms.The robustness and parallel numerical performance of the solver is investigated on large challenging linear systems arising from the finite element discretization in structural mechanics applications.     

A successive projection method for binary pattern recognition with multilayer feedforward neural networks

Error back-propagation (BP) is one of the most popular ideas used in learning algorithms for multilayer neural networks. In BP algorithms, there are two types of learning schemes, online learning and batch learning. The online BP has been applied to various problems in practice, because of its simplicity of implementation. However, efficient implementation of the online BP usually requires an ad hoc rule for determining the learning rate of the algorithm. In this paper, we propose a new learning algorithm called SPM, which is derived from the successive projection method for solving a system of nonlinear inequalities. Although SPM can be regarded as a modification of online BP, the former algorithm determines the learning rate (step-size) adoptively based on the output for each input pattern. SPM may also be considered a modification of the globally guided back-propagation (GGBP) proposed by Tang and Koehler. Although no theoretical proof of the convergence for SPM is given, some simulation results on pattern classification problems indicate that SPM is more effective and robust than the standard online BP and GGBP     

一种鲁棒BP算法及其在非线性动态系统辨识中的应用

利用多层前馈神经网络的非线性建模特性，基于动态ＢＰ网络的串并联和并联模型，提出了一种高鲁棒性ＢＰ算法，与传统的ＢＰ算法相比，鲁棒ＢＰ算法有５个优点：（１）适合于非线性动态系统辨识，（２）辨识精度高；（３）不必内插所有训练样本；（４）具有高鲁棒性，能抵制过失误差和量测误差；（５）收敛速度得到了改进，因为错误差样本的影响得到了适度的抑制，把该算法用于非线性动态系统辨识，仿真结果表明此方法是有效的。     

Explicit nonlinear dynamic finite element analysis on homogeneous/heterogeneous parallel computing environment

This paper presents parallel computational strategies to implement explicit nonlinear finite element analysis code onto distributed memory parallel computers for solving large-scale problems in structural dynamics. Implementation details on both homogeneous and heterogeneous parallel processing environments are considered in detail in this paper. Implementation of an explicit nonlinear finite element dynamic analysis code on homogeneous systems is discussed first and this is later moved onto heterogeneous systems. Domain decomposition with explicit message passing is preferred for parallel implementation. The message passing implementation in the parallel algorithm is based on MPI (Message Passing Interface) libraries. Implementation aspects of overlapped, non-overlapped domain decomposition techniques, Dynamic Task Allocation (DTA) and clustering techniques for DTA and their relative merits are presented. The interprocessor communications are optimised by overlapping with computations to improve the performance of the domain decomposition based explicit dynamic analysis finite element code.The issues related to implementation of finite element code for nonlinear dynamic analysis on heterogeneous parallel computing environment are later presented. A new dynamic load-balancing algorithm is developed for this purpose and it is integrated with the domain decomposition based parallel explicit finite element code to test our algorithms on a coarse grain heterogeneous cluster of workstations. Numerical experiments have been carried out on PARAM-10000, an Indian parallel computer and also on cluster of Unix workstations.     

GPU加速的神经网络BP算法*

近年来图形处理器(GPU)快速拓展的可编程性能力加上渲染流水线的高速度及并行性,使得图形处理器通用计算(GPGPU)迅速成为一个研究热点。针对大规模神经网络BP算法效率低下问题,提出了一种GPU加速的神经网络BP算法。将BP网络的前向计算、反向学习转换为GPU纹理的渲染过程,从而利用GPU强大的浮点运算能力和高度并行的计算特性对BP算法进行求解。实验结果表明,在保证求解结果准确度不变的情况下,该方法运行效率有明显的提高。     

基于Jini的集群网格并行计算平台的构建

研究如何使用分布式系统Jini体系来构建集群网格并行计算平台。首先介绍集群计算和Jini技术；然后给出基于Jini的集群计算系统模型JCCM，设计了一个在此模型上实现并行计算的通用算法，并在自强3000集群主机上对此模型及算法进行了测试，证实了该模型的可伸缩性和健壮性。