改进的粒子群算法

针对基本粒子群算法容易陷入局部最优点,进化后期速度慢等缺点,设计了一种新的粒子群算法,将基本粒子群算法粒子行为基于个体极值点转化为个体自身极值与其他某一个个体极值的加权平均值,而全局极值点转化为群体中优秀个体极值的加权平均值。数值仿真实验表明,新算法比PSO具有更好的收敛性,能更快地找到问题的最优解。     

基于观测的POMDP 优化算法及其仿真

在分析马尔可夫决策过程（Markov Decision Process, MDP）性能灵敏度的基础上,讨论了部分可观 测马尔可夫决策过程（Partially Observable Markov Decision Process, POMDP）的性能优化问题．给出了POMDP 性能灵敏度分析公式,并以此为基础提出了两种基于观测的POMDP 优化算法：策略梯度优化算法和策略迭 代优化算法．最后以准许控制问题为仿真实例,验证了这两个算法的有效性．     

基于FLDA、CPCA与HMM的人脸识别

为了获得具有较高识别率的算法,提出了一种将Fisher线性鉴别分析(Fisher Linear Discriminant Analysis)、复主分量分析(Principal Analysis in the Complex Space)与隐马尔可夫模型(Hidden Markov Models)相结合进行人脸识别的方法。对于输入的不同光照、人脸表情和姿势的图像先进行归一化处理,然后将归一化后的图像转化成一维向量,再用FLDA方法提取每幅图像的特征,形成新的复向量空间;通过运用复主分量分析,来抽取人脸图像的有效鉴别特征;最后通过HMM对这些特征进行训练,得到一个优化的HMM并应用于识别。在ORL人脸数据库中进行实验,实验结果表明,该方法具有较高的识别率。     

基于HMM的汉语整句拼音输入法研究

介绍了一种基于HMM的汉语整句拼音输入转换为整句汉字的输入法,提出了引入语言知识后的一种音字选择方法.并给出了采用N元拼音文法时的选择模型.实验表明,该方法取得了较好的效果.     

基于改进的Markov模型预测准确度研究

马尔可夫（Markov）模型方法主要解决的问题是从当前状态预测下一个状态,因此适合作为一个预测模型来挖掘用户的页面访问模式。此文在总结以往Markov模型方法基础上,从页面访问时间和用户个体特征两个角度对模型方法进一步改进,以提高模型预测的准确度。最后将改进的Markov方法与协同过滤方法进行了比较,指出其在计算开销和预测准确度上的优势。     

基于关联规则和序列模式挖掘的客户行为模型

首先介绍了关联规则和序列模式,并在证券交易历史数据仓库中使用Apriori算法和Aprioriall算法进行挖掘,然后用挖掘结果构造了证券网上交易的客户行为模型.最后,从该模型的Markov链转移概率矩阵出发实现了网上交易行情自动推送机制,有效地提高客户网上交易实时行情的响应速度.     

A mathematical programming approach to a problem in variance penalised Markov decision processes

In this paper we present three algorithms for solving a problem in which it is required to get an optimal compromise between the average expected reward per unit time and the variance of the reward per unit time. The algorithms lead to an-optimal solution, where>0 is arbitrary.     

CoFEE: Context-aware futuristic energy estimation model for sensor nodes using Markov model and autoregression

Nowadays, the emerging internet of things (IoT) technology offers the connectivity and communication between all things (various objects/things, devices, actuators, sensors, and mobile devices) at anywhere and anytime. These devices have embedded environment monitoring capabilities (sensors) and significant computational responsibilities. Most of the devices are working by utilizing their limited resources such as energy, memory, and bandwidth. Obviously, battery power is a crucial factor in any network. It makes tedious overheads to the network operations. Prediction of the future energy of the devices could be more helpful for managing resources, connectivity, and communication between the devices in IoT and wireless sensor networks (WSNs). It also facilitates the reliable internet and network connection establishment to the nodes. Hence, this paper presents an energy estimation model to predict the future energy of devices using the Markov and autoregression model. The proposed model facilitates smarter energy management among internet-connected devices. Performance results show that the proposed method gives significant improvement compared with the neural network and other existing predictions. Further, the proposed model has very lower error performance metrics such as mean square error and computation overhead. The proposed model yields more perfect energy predictions for a node with 64% to 97% and 16% to 43% of higher prediction accuracy throughout the time series.     

Extended state observer-based finite-region control for 2-D Markov jump systems

In this article, an extended state observer-based finite-region control scheme is presented for two-dimensional Markov jump systems with unknown mismatched disturbances. The mathematical model of the two-dimensional Markov jump systems is built on the well-known Roesser model. By establishing special recursive formulas and utilizing the 2-D Lyapunov function theory, sufficient conditions are obtained, which prove that the resultant system is finite-region bounded, if some linear matrix inequalities are achieved. Then, we provide an algorithm to solve the extended state observer-based controller gains. With the proposed control scheme, the external disturbances can be actively rejected from the system outputs. To conclude, a numerical example based on the Darboux equation is provided to demonstrate the validity and effectiveness of the devised control scheme.