一种用于多峰函数优化的改进混合遗传算法

针对遗传算法在处理复杂多峰函数优化问题中存在的一些缺陷，提出了一种改进型的混合遗传算法，它在小生境技术的基础上引入单纯搜索算法、最优保存算法和近优淘汰算法，并使之相互结合，经编程实践证明，这种改进的混合遗传算法在处理复杂多峰函数优化问题时，局部搜索能力和克服过早收敛能力方面都显著优于标准遗传算法，并在实际应用中取得了较好的效果。     

一维抛物型方程参数识别反问题的数值解法

以函数逼近和Tikhonov正则化为基础，利用算子识别摄动法和线性化技术提出求解一维抛物型偏微分方程参数识别反问题的迭代算法，拓宽了求解此类反问题泛定方程和初边值条件的适用范围。数值模拟的结果表明，用此迭代法求解参数识别反问题具有数值精度高、稳定性好、收敛速度快的特点。     

An iterative algorithm in potential-field inversion

1　Principle and MethodAll the problems of potential- field inversioncan be concluded that of solving nonlinear functionas follows:b =f( x) ( 1 )where x =( x1,x2 ,… .xn) T,xj( j =1 ,2 ,… ,n)indicate geometricparameters and physical parame-ters of the object;b=( b1,b2 ,… ,bm) T,bi( i=1 ,2 ,… ,m) indicate potential- field measurements.Main algorithms for expression( 1 ) are leastsquare method ( include damping least squaremethod ) ,gradient method ( conjugate gradientmethod and variable met…     

基于BP神经网络的直流电机PID控制系统研究

以直流电动机为控制对象,建立数学模型,并在传统PID的基础上结合神经网络算法,充分利用神经网络自学习功能,实现对PID参数的实时在线自整定.克服了PID控制参数难以确定和控制过程无法随环境等变化而自适应的缺点,体现了神经网络较好的智能性与较强的鲁棒性.利用Matlab软件进行仿真研究,结果表明神经网络PID较传统PID更精准,更具适应性,控制效果优越.     

基于A*算法的远程导弹三维航迹规划算法

提出了一种基于A*算法的三维航迹规划方法.该方法将远程导弹的运动学模型离散化,利用导弹过载作为控制量,并结合约束条件,直接对三维空间进行最优航迹搜索.航迹规划过程不仅结合了导弹的本身机动特性,并且充分利用地形信息进行地形回避和威胁回避.仿真实验表明,该方法能够直接规划出三维空间中满足导弹过载约束的最优航迹.     

混沌高效遗传算法在水库含沙量预报中的应用

利用混沌映射的遍历性和实编码遗传算法的全局优化性,通过在遗传进化过程中加入混沌变异操作,在变量的定义域内投放大量的混沌初始群体,在实编码遗传算法进化过程中加入单纯形法学习算子,建立了一种新的混沌高效遗传算法(chaos higher efficient genetic algorithm, CHEGA).应用该法对3个非线性、高维、多峰值测试函数进行了仿真,在收敛速度和全局优化方面好于现有的简单遗传算法和改进的遗传算法.建立了水库含沙量预报模型.并将CHEGA用于求解上述模型的参数优化问题,与实数编码加速遗传算法(RAGA)、二进制加速遗传算法和随机优化算法等方法相比,CHEGA可以遍历到整个区域,较好的保持了种群的多样性,并且精度高、收敛速度快.CHEGA对求解实际水库计算模型的参数优化问题非常有效.     

Reconstruction of linkage maps in the distorted segregation populations of backcross, doubled haploid and recombinant inbred lines

Non-Mendelian segregation of markers, known as distorted segregation, is a common biological phenomenon. Although segregation distortion affects the estimation of map distances and the results of quantitative trait loci (QTL) mapping, the effects of distorted markers are often ignored in the construction of linkage maps and in QTL mapping. Recently, we have developed a multipoint method via a Hidden Markov chain method to reconstruct linkage maps in an F2 population that corrects for bias of map distances between distorted markers. In this article, the method is extended to cover backcross, doubled haploid and recombinant inbred line (RIL) populations. The results from simulated experiments show that: (1) the degree that two linked segregation distortion loci (SDL) affect the estimation of map distances increases as SDL heritability and interval length between adjacent markers increase, whereas sample size has little effect on the bias; (2) two linked SDL result in the underesti- mation of linkage distances for most cases, overestimation for an additive model with opposite additive effects, and unbiased estimation for an epistatic model with negative additive-by-additive effects; (3) the proposed method can obtain the unbiased estimation of linkage distance. This new method was applied to a rice RIL population with severely distorted segregation to reconstruct the linkage maps, and a bootstrap method was used to obtain 95% confidence intervals of map distances. The results from real data analysis further demonstrate the utility of our method, which provides a foundation for the inheritance analysis of quantitative and viability traits.     

支持向量机增量学习算法综述

支持向量机增量学习算法,有效的解决了因数据集庞大而引起的内存不足问题,改善了因出现新样本而造成原分类器分类精度降低、分类时间延长的局面。本文阐述了几种具有代表性的增量学习算法,比较了它们的优缺点,给出了进一步的研究方向。     

捆用工资实施的细化与探索

从标准公开、程序公开、考核公开及结果公开等方面探讨了捆用工资实施的作用与效果。     

一种基于Rough sets生成决策树的算法改进

介绍了决策树算法的含义和构筑方法，对基于加权平均粗糙度构造决策树算法进行改进，通过实例说明了改进算法的优势。