基于高斯过程回归和双滑动窗口残差处理的风电机组主轴状态监测

针对风电机组运行数据强随机性和高噪声的特点,采用高斯过程回归方法建立主轴正常时段的温度模型,并应用自动关联分析方法选择主轴温度模型的建模变量。为降低建模复杂程度,采用模糊核聚类方法对风电机组原始运行数据进行筛选,消除冗余信息,构造紧凑有效的建模样本集。当主轴发生故障时,模型的输入观测向量发生异常变化,导致模型预测残差发生明显改变。为提高主轴异常预警的灵敏度和可靠性,采用基于莱依特准则的双滑动窗口对预测残差序列进行实时的统计分析,如果残差均值或标准差超出设定的故障报警阈值,则发出报警信息。某风电机组主轴的实际运行数据验证了所提方法的有效性。

一种基于DGMF与PSO的多核SVM

目前常用的支持向量机采用传统的单一高斯核函数,或者传统高斯核函数与多项式核函数的组合,但是存在着分类器泛化推广能力不强,容易出现数据偏斜等问题。鉴于此,提出了基于DGMF的多核支持向量机、并采用粒子群算法优化其参数。最后构造了文本倾向性分类实验,实验结果表明改进算法在正确率、准确率、召回率、F1值上具有更好的性能。

Design of a Tree Ring Structure Analysis System to Estimate the Accurate Age of Tree Species in Sri Lanka

Determination of an age in a particular tree species can be considered as a vital factor in forest management.In this research we have introduced a novel scheme to determine the accurate age of the tree species in Sri Lanka.This is initially developed for the tree species called‘Hora’(Dipterocarpus zeylanicus)in wet zone of Sri Lanka.Here the core samples are extracted and further analyzed by means of the different image processing techniques such as Gaussian kernel blurring,use of Sobel filters,double threshold analysis,Hough line tran sformation and etc.The operations such as rescaling,slicing and measuring are also used in line with image processing techniques to achieve the desired results.Ultimately a Graphical user interface(GUI)is developed to cater for the user requirements in a user friendly environment.It has been found that the average growth ring identification accuracy of the proposed system is 93%and the overall average accuracy of detecting the age is 81%.Ultimately the proposed system will provide an insight and contributes to the forestry related activities and researches in Sri Lanka.

基于M^3-DGMF的债券评级算法研究

KNN、SVM等算法已经较为普遍地使用在信用评级中,但债券评级数据有着数量庞大、样本向量不一致、多分量等特征,经典算法在解决这种问题时往往局限性较大。采取双高斯合成函数的最小最大模块化神经网络的监督聚类算法对债券评级数据进行学习、泛化,实验结果表明该算法具有较大优势。