一种非线性降维算法在组合预测模型中的应用*

针对视频序列维数高、帧间相关性大、运动轨迹复杂的特点,将LLE非线性降维算法用于视频处理,并重点研究了如何利用该算法对目标跟踪过程中的模板进行预测更新.由于单步预测方法在运动目标发生部分或全部遮挡时无法保证跟踪的准确性,进一步将时间序列模型与BP网络相结合实现跟踪目标的多步预测,从而可以弥补时间序列模型在单步预测方面的不足.实验证明,该算法能保证在运动目标跟踪过程中的准确性和鲁棒性.     

一种降维新方法*

针对图像数据的72维HSV颜色特征,对数据集的本征维概念进行了有意义的扩展,在此基础上提出了一个新的降维机制.实验证明,该降维算法是行之有效的.     

基于非线性降维算法的视频序列特征提取及图像重建

在经典算法的基础上,根据典型视频序列的特点对降维算法进行双通道扩展,并提出k近邻核函数法重建高维图像.在高低维空间建立映射,解决无重构算法的问题.同时也为视频压缩提供一条新思路.实验结果显示算法的有效性.     

非线性降维算法及其在医院绩效考核上的应用

流形学习算法中的等距嵌入算法（ISOMAP）具有对离群点敏感的瑕疵,针对此问题,提出利用基于共享近邻的距离度量方式,并充分利用了流形上对象的局部密度信息,有效改善了算法的性能,提高了算法的健壮性。同时,首次尝试将该改进的流形学习算法应用于医院绩效考核。人工数据与真实数据上的实验表明,改进的算法健壮且有效,在绩效考核上应用成功。     

一种混合量子粒子群算法在神经网络设计中的应用

将混合量子粒子群算法（HQPSO）应用于神经网络设计,可以在对网络拓扑结构优化的同时对连接权重进行求解。该算法引入了选择机制,使优势粒子得以保留,并在训练后期使用BP算法提高训练精度,具有较高的进化效率。通过对混沌时序信号的预测,表明HQPSO算法改进了神经网络的学习性能和泛化能力。     

季节性组合预测模型在医院门诊量中的应用研究

医院门诊量是一个具有复杂的非线性组合特征的季节性时问序列。为解决传统时间序列预测大多数都是使用单一模型，以致影响预测精度等问题，采用了最优加权组合预测方法将季节性ARIMA乘积模型和季节性神经网络模型进行组合优化。结果表示，季节性组合预测模型在拟合精度和预测准确性方面优于任何单一预测方法，为季节性时间序列预测提供了一种新的实用方法。     

基于非线性降维算法的膜蛋白类型识别

众所周知,研究未知膜蛋白的类型可对基础研究和药物发现提供有用的线索。在后基因组时代,伴随着蛋白质序列数量的剧增,用实验方法确定膜蛋白类型太过昂贵和费时。因此,研究出一种能够自动发现可能的膜蛋白的计算方法变得很重要。鉴于这种情况,曾有人采用DC(Dipeptide Composition)方法表示蛋白质序列并取得了很好的预测结果。然而,采用这种表示方法得到的特征维数很高,冗余很大,使得预测系统十分复杂。为了解决这个问题,本文采用非线性降维算法KPCA(Kernel Principle component analysis),通过从高维的DC(Dipeptide Composition)特征空间中提取出低维的重要特征来简化该系统,采用K-NN(K-nearest neighbor)分类器从约简后的低维特征中预测膜蛋白类型。实验结果表明,使用KPCA方法预测膜蛋白类型非常有效。     

一种在源数据稀疏情况下的数据降维算法

流形学习方法是根据流形的定义提出的一种非线性数据降维方法,主要思想是发现嵌入在高维数据空间的低维光滑流形。从分析基于流形学习理论的局部线性嵌入算法入手,针对传统的局部线性嵌入算法在源数据稀疏时会失效的缺点,提出了基于局部线性逼近思想的流形学习算法,并在S-曲线上采样测试取得良好降维效果。     

基于神经网络模型的时间序列预测算法及其应用

提出了一种神经网络模型的时间序列直接多步预测算法。网络的学习采用具有遗忘因子的ＢＰ算法与时差方法相结合的混合算法，解决了经典ＢＰ算法在直接多步预测中不能渐进计算的问题，同时网络具备一定的结构学习能力。采用该算法对现场采集的高炉铁水含硅量时间序列数据进行预报实验，表明本文提出的直接多步预测方法是可行的。     

一种监督降维的OP-LASRC算法在行为识别中的应用*

针对大数据的人体行为识别时实时性差和识别率低的问题,提出了优化投影对线性近似稀疏表示分类(OP-LASRC)的监督降维算法。OP-LASRC将高维的行为数据优化投影到低维空间,与线性近似稀疏表示(LASCR)快速分类算法相结合应用大数据的人体行为识别。首先利用LASCR的残差计算规律设计OP-LASRC算法,实现监督降维;利用线性正交投影缩减高维数据的维度,投影时减小训练样本的本类重构残差及增大类间重构残差,从而保留训练样本的类别特征。然后,对降维后的行为数据,利用LASCR算法进行分类;用L2范数估算稀疏系数,选出前k个最大的稀疏系数对应的训练样本,缩减训练样本库后用L1范数最小化和残差最小化计算得到识别结果,从识别率、鲁棒性、执行时间三个方评价此方法,在KTH行为数据库上进行实验测试。实验表明：OP-LASRC监督降维后,LASRC在分类时不仅识别率高达96.5%,执行时间比同类算法短,而且保证了强鲁棒性,证明了OP-LASRC能完美匹配LASCR算法用于行为识别,这为大数据的行为识别提供了一种新的思路。     

基于遗传算法的非线性模型预测控制方法

介绍了非线性模型预调控制算法结构，提出了基于遗传算法的非线性模型预测控制方法，将遗传算法作为优化技术用于受限非线性模型预测控制器的设计。算法采用双模控制策略，将保证预测控制算法稳定性的终点等式约束转化为终点不等式约束，以利于遗传算法的实施。基于不变集理论，给出了非线性模型预测控制算法的稳定性定理。仿真结果表明了所提出控制算法的可行性和有效性。     

An iterative model predictive control algorithm for constrained nonlinear systems

An iterative model predictive control (MPC) scheme for constrained nonlinear systems is presented. The idea of the method is to detour from the solution of a non‐convex optimization problem using a time‐variant linearization of the nonlinear system model that is adjusted iteratively by solving an iterative quadratic programming optimization problem at each sampling time. The main advantage is the faster resolution of the optimization problem by using quadratic programming instead of non‐convex programming and yet, properly describing the nonlinear dynamics of the process being controlled. In this article, a general framework of the method is presented together with a discussion on the conditions under which the iterations converge and on the uncertainty of its results due to the linearization used, as well as some practical considerations about its implementation. The performance of the proposed controller is illustrated via two examples.     

Application of nonlinear rescaling method to model predictive control

In this paper, we present the application of a nonlinear rescaling method for an efficient solution to the constrained optimal control problem in model predictive control. A smooth scaling function is used to transform the constraints to a sequential unconstrained optimization problem. Simulation results demonstrate the effectiveness of this approach with regards to model predictive controller with saturation.     

基于组合模型的非线性时间序列预测算法

为解决复杂系统中非线性时间序列预测模型构建效率低、预测精度低的问题,提出基于组合模型的HURST-EMD预测算法.采用EMD算法将非线性时间序列分解为代表原始序列特征的各个IMF,然后引入赫斯特(Hurst)指数将同类的IMF整合为新的分量,最后选用LS-SVR-ARIMA模型进行组合预测.在该算法中,设计了序列分类整合等过程,优化了建模的计算量,构建了高效精准的预测模型.为验证模型的有效性,采用上证指数公共数据集和真实交通流数据进行检验,实验结果表明,改进的基于组合模型的HURST-EMD预测算法在提高预测效率的同时具有更好的预测精度.     

Advanced-multi-step nonlinear model predictive control

Nonlinear model predictive control (NMPC) has gained widespread attention due to its ability to handle variable bounds and deal with multi-input, multi-output systems. However, it is susceptible to computational delay, especially when the solution time of the nonlinear programming (NLP) problem exceeds the sampling time. In this paper we propose a fast NMPC method based on NLP sensitivity, called advanced-multi-step NMPC (amsNMPC). Two variants of this method are developed, the parallel approach and the serial approach. For the amsNMPC method, NLP problems are solved in background multiple sampling times in advance, and manipulated variables are updated on-line when the actual states are available. We present case studies about a continuous stirred tank reactor (CSTR) and a distillation column to show the performance of amsNMPC. Nominal stability properties are also analyzed.     

Estimation and adaptive nonlinear model predictive control of selective catalytic reduction systems in automotive applications

The demands of high NO_x conversion efficiency and low tailpipe ammonia slip for urea-based selective catalytic reduction (SCR) systems have been substantially increased in the past decade, as NO_x emission legislations for Diesel engines are becoming more stringent than ever before. Since catalyst aging has a significant impact on SCR performance, robust and adaptive SCR control has been preferred for degraded SCR systems to realize emission control objectives. The purpose of this paper is twofold. Firstly, a robust ammonia coverage ratio observer was designed for estimating the ammonia coverage ratio reference for catalysts with different aging levels. An ammonia storage capacity observer was developed for estimating the actual ammonia storage capacity which can be reduced due to catalyst aging. An adaptive ammonia coverage ratio reference design was then developed to estimate the desired ammonia coverage ratio ranges at each instantaneous engine operating point for both single-cell and two-cell SCR systems at different aging levels based on a singular perturbation method. Secondly, to ensure the estimated ammonia coverage ratio falls in the desired ranges for most of engine operating conditions, robust nonlinear model predictive control (NMPC) algorithms were designed for both single-cell and two-cell SCR systems. Experimental data over US06 cycle were collected from a Diesel engine and aftertreatment system platform for controller verification. Simulation results under US06 test cycle demonstrate that the proposed NMPC algorithms were capable of consistently achieving high NO_x conversion efficiency (>95.6%) and constrained tailpipe ammonia slip (<10 ppm on average and <12 ppm on the peak) for both fresh catalyst and aged catalyst with 30% loss of ammonia storage capacity.     

Structure-preserving infinite dimensional model reduction: Application to adsorption processes

This paper deals with the spatial discretization of distributed parameter systems. The originality of the proposed approach is to combine geometrical modelling and finite element discretization method to preserve the model structure associated with both mass and energy balances during the reduction. The approach is presented through the example of an adsorption process whose main feature is that it is an heterogeneous system. The methodology is described on the microporous phase and simulations are carried out on the full multilevel adsorption column.     

A robust model predictive control algorithm for incrementally conic uncertain/nonlinear systems

This paper presents a robustly stabilizing model predictive control algorithm for systems with incrementally conic uncertain/nonlinear terms and bounded disturbances. The resulting control input consists of feedforward and feedback components. The feedforward control generates a nominal trajectory from online solution of a finite‐horizon constrained optimal control problem for a nominal system model. The feedback control policy is designed off‐line by utilizing a model of the uncertainty/nonlinearity and establishes invariant ‘state tubes’ around the nominal system trajectories. The entire controller is shown to be robustly stabilizing with a region of attraction composed of the initial states for which the finite‐horizon constrained optimal control problem is feasible for the nominal system. Synthesis of the feedback control policy involves solution of linear matrix inequalities. An illustrative numerical example is provided to demonstrate the control design and the resulting closed‐loop system performance. Copyright © 2010 John Wiley & Sons, Ltd.