自适应的多种群并行遗传算法研究

典型的遗传算法本质上是一种并行的随机搜索策略,它不能对进化方向做出正确的感知和预测。文章定义了一个可以感知进化方向和衡量进化速度的指标,指导遗传算子做出自适应的调整。结合并行计算和多种群进化思想,提出了“带环多种群模型”,并构造了一种自适应的多种群并行遗传算法。实验结果表明,该算法可以有效地引导和控制进化方向,克服过早收敛现象,提高搜索效率。     

基于协同演化遗传算法的文本特征获取方法

信息过滤技术是解决“信息过载”和“信息迷向”问题的有效手段。为高效地确立用户的信息需求模型,提出利用协同演化的遗传算法解决多主题多文本的特征获取问题。协同演化遗传算法根据种群中个体进化速度、效果的不同,采取相互评价、相互学习、群体进化的协同演化策略,使得个体在其它种群、个体的指导下,不断获得较好的基因,从而实现文本特征的抽取。实验验证了方法的有效性。     

基于协同进化遗传算法的自动谈判

提出了基于协同进化遗传算法的自动谈判算法,模拟了有限期轮流出价谈判协议中的策略学习机制。实验结果表明,基于协同进化的自动谈判能够生成近似于子博弈完美均衡的策略组合,具有良好的应用前景。 自动谈判;协同进化;遗传算法;轮流出价     

一种基于混沌迁移的伪并行遗传算法及其应用

为了解决遗传算法寻优过程中的早熟收敛问题 ,本文提出了一种基于混沌迁移策略的伪并行遗传算法 ,该算法针对实时性要求不高的优化问题采用串行的算法结构实现分解型并行遗传算法的“独立进化、信息交换”思想 .在并行进化的个体异步迁移过程中 ,引入了混沌迁移序列引导个体迁移过程 ,利用其遍历性和随机性 ,保证了子种群之间能够进行充分高效的信息交换 .仿真研究和在库存优化方面的应用研究表明 ,这种算法具有很强的全局搜索能力 ,寻优效率高 ,有效克服了标准遗传算法的早熟收敛问题 .     

改进的遗传优化BP-NN在变电站故障区域中的研究

BP神经网络（Back Propagation Neural Network,BP-NN）具有良好的自学习能力以及自适应和泛化能力,但运算过程中容易陷入局部极小值,同时隐含层节点数的选择也影响着诊断的效果。文中根据经验公式缩小隐层节点数范围,在小范围里寻找最优的隐层节点数。根据遗传算法(Genetic Algorithm,GA)具有全局寻优的特点,用遗传算法优化BP神经网络训练的初始权值阈值,可以避免BP神经网络陷入局部极小的问题。但是,传统遗传算法也有自身的缺点,其在全局寻优的过程中,易陷入“早熟”的问题。为了解决传统遗传算法“早熟”现象,文中提出了一种协同进化的遗传算法,即使用3个种群同时进化的遗传算法,协同进化遗传算法不但可以避免传统遗传算法的“早熟”问题,而且可以加强局部搜索提高运行效率。将协同进化遗传算法应用到BP神经网络中,仿真结果表明,该方法可以准确有效地诊断出变电站故障元件,提高变电站故障诊断过程中的容错性及效果。     

一种改进的抑制早熟收敛的模糊遗传算法

针对遗传算法中的早熟收敛现象,提出一种改进的模糊遗传算法。该算法将群体适应度均方差和种群的进化代数作为模糊逻辑控制器判断早熟收敛的标准,并根据判断结果对优劣不等的个体采取相应的进化方法,即当种群正常进化时对个体执行“惩强扶弱”的措施以保持种群多样性,一旦发生早熟收敛或有早熟收敛的趋势则对劣质个体进行局部灾变,以恢复种群的进化能力。实验结果表明,与标准遗传算法、自适应遗传算法和模糊遗传算法相比,改进的模糊遗传算法能够更好地维持种群多样性,抑制早熟收敛。     

遗传算法在人工智能领域中的应用

该文介绍了遗传算法的基本概念、基本遗传算法的特点和基本遗传算法的求解步骤,同时也介绍了遗传算法在机器学习、并行处理、人工生命以及遗传算法与进化规则及进化策略的结合的发展动向,最后讨论了基于遗传算法的人工神经网络学习中的应用研究,具体论述了遗传算法在学习神经网络权重和学习神经网络拓扑结构的应用方法。     

基于协同进化的自适应遗传算法研究

针对传统遗传算法易于陷入局部最优解,性能不稳定的问题,提出了一种基于协同进化的自适应遗传算法（CEAGA）。在协同进化的两层框架模型的基础上,引入一个自适应的变异策略,改进了协同进化遗传算法中的局部进化操作,加强了在上层中的局部搜索;在下层,在种群之间采用协同进化算法,克服未成熟收敛,在种群内部进化中引入自适应遗传操作,保护种群中的优秀个体。实验验证CEAGA既具有很快的收敛速度,又具有很好的全局搜索性能。     

基于多种编码的多群体遗传算法

为了有效地克服标准遗传算法（SGA）中的早熟收敛现象,提出了一种基于多种编码的多群体遗传算法,该方法是采用3个群体同时进行进化的策略,其中,第1个 本是采用浮点数编码方法,以使该群体具有较强的局部搜索能力,第2个群体是采用二进制编码方法,以使该群体具有较强的全局搜索能力。第3个群体为“精华种群”,用于保存算法在进化过程中产生的优秀个体,在进化过程中,还通过引入“移民”策略来交换3个群体中的优秀个体,以有效地增加群体的多样性,该算法不仅不易陷入局部收敛,还具有较强的跳出局部收敛的能力,且收敛速度较快,通过对一系列典型复杂多模函数进行的优化计算试验,结果证实了该方法的有效性和优越性。     

一种多精英保存策略的遗传算法

根据种子到当前最优点的距离将种群分成两部分,小于或等于某一自适应距离值的种子归入当前最优种群,大于该距离值的次优种子形成次优种群集合。对此两个种群分别按照不同的进化策略协同进化并重组。通过界定最优种群边界来提高遗传算法局部搜索能力,通过对次优种群自适应变异,比较好地平衡种群的“选择压力”和“种群多样性”。数值结果表明了本方法的有效性和稳定性。     

Global replacement-based differential evolution with neighbor-based memory for dynamic optimization

Dynamic optimization problems challenge the evolutionary algorithms, owing to the diversity loss or the low search efficiency of the algorithms, especially when the problems change frequently. This paper presents a novel differential evolution algorithm to address the dynamic optimization problems. Unlike the most used “DE/rand/1” mutation operator, in this paper, the “DE/best/1” mutation is employed to generate a mutant individual. In order to enhance the search efficiency of differential evolution, the classical differential evolution algorithm is modified by a novel replacement operator, in which the worst individual in the whole population is replaced by the newly generated trial vector as a “steady-state” manner. During optimizing, some newly generated solutions are stored into a memory set, in which these stored solutions are located around the current best solution. When the environmental change is detected, the stored solutions are expected to guide the reinitialized solutions to track the new location of global optimum as soon as possible. The performance of the proposed algorithm is compared with six state-of-the-art dynamic evolutionary algorithms over some benchmark problems. The experimental results show that the proposed algorithm clearly outperforms the competitors.     

自适应群体结构的粒子群优化算法

粒子群优化算法中,群体结构的组织模式直接决定了粒子间信息的共享和交流方式.根据复杂网络形成过程中的动力学原理,提出了一种自适应群体结构的粒子群优化算法.算法初期粒子空间分布分散,搜索过程中不断产生新的连接,群体的搜索模式由L_best? 模型逐渐进化为G_best? 模型,群体结构的这种进化方式有利于算法早期的“勘探”和后期的“开采”.实验结果表明,新算法在收敛性能上获得了较大提高.     

A parallel ETD algorithm for large-scale rate theory simulation

Rate theory (RT) is a commonly used method to simulate the evolution of material defects. A promising numerical method, exponential time difference (ETD), can reduce the stiff RT equations to explicit ordinary differential equations (ODEs). Previous implementations of ETD on the “Sunway TaihuLight” supercomputer suffer from high computation cost and poor parallel efficiency while solving a large amount of ODEs. This paper improves the algorithm with hybrid MPI+SIMD and additional instruction-level optimizations by taking advantage of the architecture of “Sunway TaihuLight”. The execution time of a single iteration is reduced by about 40%. Scaling from 64 to 4096 processes, the parallel efficiency of the new algorithm achieves 33.5% and 50.6% in strong and weak scalability, which corresponds to 21.4 and 32.4 in speedup, respectively.

     

Uniform random generation of expressions respecting algebraic identities

Algorithms for the uniform random generation of a particular class of formal expressions (containing arithmetical expressions, propositional calculus formulas, tree representations, special algebraic expressions and program structures) are described. “Uniform” means that all non-equivalent expressions of the same size are equiprobable, where equivalence is induced by commutative or associative properties of certain symbols (e.g.“a+b”≡“b+a”). In the special case where no commutative symbols occur, it is shown that the problem can be treated by a modification of Hickey's and Cohen's well known generation algorithm for context-free languages. In order to obtain a speed up in the generation time, a new, parallelizable algorithm is developed, which turns out to be applicable also to the general case (occurrence of commutative symbols).     

Partial adiabatic quantum search algorithm and its extensions

In this paper, we again discuss quantum search by partial adiabatic evolution, which was first proposed by Zhang et al. In contrast to previous conclusions, we show that partial adiabatic search does not improve the time complexity of a local adiabatic algorithm. Firstly, we show a variant of this algorithm and find that it is equivalent to the original partial adiabatic algorithm, in the sense of the same time complexity. But we give two alternate viewpoints on this “new” adiabatic algorithm—“global” adiabatic evolution and local adiabatic evolution approaches, respectively. Then, we discuss how global and local adiabatic quantum search can be recast in the framework of partial adiabatic search algorithm. It is found here that the former two algorithms could be considered as special cases of the later one when appropriately tuning the evolution interval of it. Also this implies the flexibility of quantum search based on partial adiabatic evolution.     

An antigen space density based real-value negative selection algorithm

The negative selection algorithm (NSA) is an important detector generation algorithm for artificial immune systems. In high-dimensional space, antigens (data samples) distribute sparsely and unevenly, and most of them reside in low-dimensional subspaces. Therefore, traditional NSAs, which randomly generate detectors without considering the distribution of the antigens, cannot effectively distinguish them. To overcome this limitation, the antigen space density based real-value NSA (ASD-RNSA) is proposed in this paper. The ASD-RNSA contains two new processes. First, in order to improve detection efficiency, ASD-RNSA utilizes the antigen space density to calculate the low-dimensional subspaces where antigens are densely gathered and directly generate detectors in these subspaces. Second, to eliminate redundant detectors and prevent the algorithm from prematurely converging in high-dimensional space, ASD-RNSA suppresses candidate detectors that are recognized by other mature detectors and adopts an “antibody suppression rate” to replace the “expected coverage” as the termination condition. Experimental results show that ASD-RNSA achieves a better detection rate and has better generation quality than classical real-value NSAs.     

Parallel adaptive mesh generation

Unstructured meshes have proved to be a powerful tool for adaptive remeshing of finite element idealizations. This paper presents a transputer-based parallel algorithm for two dimensional unstructured mesh generation. A conventional mesh generation algorithm for unstructured meshes is reviewed by the authors, and some program modules of sequential C source code are given. The concept of adaptivity in the finite element method is discussed to establish the connection between unstructured mesh generation and adaptive remeshing.After these primary concepts of unstructured mesh generation and adaptivity have been presented, the scope of the paper is widened to include parallel processing for un-structured mesh generation. The hardware and software used is described and the parallel algorithms are discussed. The Parallel C environment for processor farming is described with reference to the mesh generation problem. The existence of inherent parallelism within the sequential algorithm is identified and a parallel scheme for unstructured mesh generation is formulated. The key parts of the source code for the parallel mesh generation algorithm are given and discussed. Numerical examples giving run times and the consequent “speed-ups” for the parallel code when executed on various numbers of transputers are given. Comparisons between sequential and parallel codes are also given. The “speed-ups” achieved when compared with the sequential code are significant. The “speed-ups” achieved when networking further transputers is not always sustained. It is demonstrated that the consequent “speed-up” depends on parameters relating to the size of the problem.     

Combining biased randomization with iterated local search for solving the multidepot vehicle routing problem

This paper proposes a hybrid approach for solving the multidepot vehicle routing problem (MDVRP) with a limited number of identical vehicles per depot. Our approach, which only uses a few parameters, combines “biased randomization”—use of nonsymmetric probability distributions to generate randomness—with the iterated local search (ILS) metaheuristic. Two biased‐randomized processes are employed at different stages of the ILS framework in order to (a) assign customers to depots following a randomized priority criterion—this allows for fast generation of alternative allocation maps and (b) improving routing solutions associated with a “promising” allocation map—this is done by randomizing the classical savings heuristic. These biased‐randomized processes rely on the use of the geometric probability distribution, which is characterized by a single and bounded parameter. Being an approach with few parameters, our algorithm does not require troublesome fine‐tuning processes, which tend to be time consuming. Using standard benchmarks, the computational experiments show the efficiency of the proposed algorithm. Despite its hybrid nature, our approach is relatively easy to implement and can be parallelized in a very natural way, which makes it an interesting alternative for practical applications of the MDVRP.     

基于BBS产生器和椭圆曲线的改进RC4算法

针对RC4算法密钥流序列随机性不高,易受故障引入攻击、区分攻击和“受戒礼攻击”的问题,提出了一种基于BBS产生器和椭圆曲线的RC4改进算法。该算法利用随机比特产生器和随机大素数生成种子密钥Key,利用椭圆曲线产生秘密整数,在每次输出后对S盒中元素重新赋值,生成随机性很高的密钥流序列。改进RC4算法可以通过NIST随机性测试,其中频率检验、游程检验和Maurer检验等比RC4算法分别高出0.129 18,0.107 39,0.197 64,能够有效防止不变性弱密钥的产生,抵抗“受戒礼”攻击;密钥流序列分布均匀,不存在偏差,能够有效抵御区分攻击;基于椭圆曲线产生的秘密整数猜测困难,S盒内部状态不能获知,能够抵抗“故障引入”攻击。理论和实验证明改进RC4算法的随机性和安全性高于RC4算法。