Self-adaptation of learning rate in XCS working in noisy and dynamic environments

An extended classifier system (XCS) is an adaptive rule-based technique that uses evolutionary search and reinforcement learning to evolve complete, accurate, and maximally general payoff map of an environment. The payoff map is represented by a set of condition-action rules called classifiers. Despite this insight, till now parameter-setting problem associated with LCS/XCS has important drawbacks. Moreover, the optimal values of some parameters are strongly influenced by properties of the environment like its complexity, changeability, and the level of noise. The aim of this paper is to overcome some of these difficulties by a self-adaptation of a learning rate parameter, which plays a key role in reinforcement learning, since it is used for updates of classifier parameters: prediction, prediction error, fitness, and action set estimation. Self-adaptive control of prediction learning rate is investigated in the XCS, whereas the fitness and error learning rates remain fixed. Simultaneous self-adaptation of prediction learning rate and mutation rate also undergo experiments. Self-adaptive XCS solves one-step problems in noisy and dynamic environments.     

Classification-based self-adaptive differential evolution with fast and reliable convergence performance

To avoid the problems of slow and premature convergence of the differential evolution (DE) algorithm, this paper presents a new DE variant named p-ADE. It improves the convergence performance by implementing a new mutation strategy “DE/rand-to-best/pbest”, together with a classification mechanism, and controlling the parameters in a dynamic adaptive manner, where the “DE/rand-to-best/pbest” utilizes the current best solution together with the best previous solution of each individual to guide the search direction. The classification mechanism helps to balance the exploration and exploitation of individuals with different fitness characteristics, thus improving the convergence rate. Dynamic self-adaptation is beneficial for controlling the extent of variation for each individual. Also, it avoids the requirement for prior knowledge about parameter settings. Experimental results confirm the superiority of p-ADE over several existing DE variants as well as other significant evolutionary optimizers.     

朴素差分进化算法

针对变异算子学习方式的单一性,提出一种朴素变异算子,其基本思想是向优秀的个体靠近,同时远离较差个体,其实现方式是设计一种缩放因子调整策略,如果三个随机个体在某维上比较接近,则缩放因子变小,反之变大.在实验过程中通过平均适应度评价次数、成功运行次数和加速比等指标表明,基于朴素变异算子的差分进化算法能有效提高算法的收敛速度和健壮性.     

以优秀个体为导向的多策略差分进化算法

提出一种以优秀个体为导向的多策略差分进化算法.根据适应度值将种群等分为三个子种群,针对不同的种群使用不同的变异策略和控制参数.针对适应度值较差的种群提出了一种新的变异策略,通过引入学习因子和平衡因子,对提高收敛速度、精度和易陷入局部最优状态进行平衡,并对其中个体的控制参数采取自适应的机制,降低种群陷入停滞状态的概率.除...     

Cauchy mutation based on objective variable of Gaussian particle swarm optimization for parameters selection of SVM

On the basis of the slow convergence of particle swarm algorithm (PSO) during parameters selection of support vector machine (SVM), this paper proposes a hybrid mutation strategy that integrates Gaussian mutation operator and Cauchy mutation operator for PSO. The combinatorial mutation based on the fitness function value and the iterative variable is also applied to inertia weight. The results of application in parameter selection of support vector machine show the proposed PSO with hybrid mutation strategy based on Gaussian mutation and Cauchy mutation is feasible and effective, and the comparison between the method proposed in this paper and other ones is also given, which proves this method is better than sole Gaussian mutation and standard PSO.     

基于自适应变异算子的差分进化算法

针对差分演化算法易于早熟、收敛速度慢和收敛精度低等问题,提出一种基于自适应变异算子的差分进化算法。给出个体向量粒子及维度层定义,并提出了基于维度层加权的异维维度选择策略,首次将加权异维学习策略引入差分演化算法中,有效地提高了种群的多样性;根据种群聚集度的思想,提出一种基于种群聚集度自适应的变异算子,该算子能依据种群个体当前的种群聚集度自适应地调整DE/best/1变异算子和加权异维学习变异算子的变异权重,加快算法收敛速度、提高其收敛精度。通过在20个典型的测试函数上进行测试,与7种具有代表性的算法相比,结果表明提出的算法在求解精度和收敛速度上具有很大优势,并显示出了非常好的鲁棒性。     

精英化岛屿种群引导的差分进化算法

针对差分进化算法常见的早熟收敛、搜索停滞和求解精度低的问题,研究一种精英化岛屿种群的差分进化算法（EIDE）。为了实现全局搜索与局部搜索能力并重,EIDE划分多个岛屿种群,根据迭代时的适应度情况,动态地将岛屿种群分类为精英岛屿和普通岛屿;针对精英岛屿,提出一种控制参数自适应方法,依据岛屿适应度情况,自适应地调整变异概率与交叉概率,同时算法利用增强局部搜索的变异策略,提高收敛速度与精度;针对普通岛屿,使用适合全局搜索的变异与交叉概率及变异策略,维护种群多样性。EIDE提出了一种可控的“移民”与“个体迁移”策略,控制优质基因流动,有效避免早熟收敛与搜索停滞问题。在9个benchmark函数上的测试结果表明,新算法具有较强的全局寻优能力与稳定性,且收敛速度较快。     

基于变异算子和邻域值自适应的MOEA/D算法

基于分解的多目标进化算法（MOEA/D）在解决多目标问题时，具有简单有效的特点。但多数MOEA/D采用固定的控制参数，导致全局搜索能力差，难以平衡收敛性和多样性。针对以上问题提出一种基于变异算子和邻域值自适应的多目标优化算法。该算法根据种群中个体适应度值的分散或集中程度进行判断，并据此对变异算子进行自适应的调节，从而增强算法的全局搜索能力；根据进化所处的阶段以及个体适应度值的集中程度，自适应地调节邻域值大小，保证每个个体在不同的进化代数都有一个邻域值大小；在子问题邻域中，统计子问题对应个体的被支配数，通过判断被支配数是否超过设定的上限，来决定是否将Pareto支配关系也作为邻域内判断个体好坏的准则之一。将提出的算法与传统的MOEA/D在标准测试问题上进行对比。实验结果表明，提出的算法求得的解集具有更好的收敛性和多样性，在求解性能上具有一定的优势。     

带偏向性轮盘赌的多算子协同粒子群优化算法

针对粒子群优化算法在处理高维、大规模、多变量耦合、多模态、多极值属性优化问题时易早熟收敛等性能和技术瓶颈,基于粒子群优化算法行为学习算子和3种不同学习偏好的差分变异算子,建立带偏向性轮盘赌的多算子选择与融合机制,提出一种带偏向性轮盘赌的多算子协同粒子群优化算法MOCPSO.MOCPSO针对迭代粒子群榜样粒子集,首先通过对迭代种群及其榜样粒子集优劣分组,同时采用轮盘赌分别为每组榜样粒子集选配不同学习偏好的变异算子,并为每组榜样粒子适配差分基向量和最优基向量,预学习并优化迭代种群及其榜样粒子,以权衡算法的全局探索和局部开发;然后通过合并所有子种群,并结合粒子群优化算法行为学习算子,指导迭代种群状态更新,以提高算法的全局收敛性;最后结合精英学习策略,对群体历史最优进行高斯扰动,以提高算法的局部逃生能力,保障算法收敛的多样性.实验结果表明,MOCPSO算法与5种先进的同类型群智能算法在求解CEC2014基准测试问题上具备竞争力,且有更强的优化特性.     

Artificial worlds modeling of human resource management systems

Effective human resource management facilitates the success of an organization and the progress of a society. We describe an evolutionary computer model that simulates different modes of interaction between people and their environment. A two-level genotype-phenotype structure is used to represent the characteristics of an individual. The environment is modeled as a two-dimensional array of regions in which each region is characterized by a set of regional features and organizational culture. Evolution can occur at the regional and organizational levels. At the level of regional learning, the experimental results show that people tend to migrate from lesser-fitting regions to better-fitting regions to increase their fitness, which in turn results in the problem that some regions become extremely crowded and other areas have few residents. This problem can be partially eased by putting pressure on the number of people allowed in each region. However, our results show that too great an increase in pressure worsens the problem. At the level of organizational learning, our experiments show that individuals with a local mutation operator are better at adapting to a constant leadership strategy (type), while those with a global mutation operator are better at coping with the changes in leadership strategy. The individuals who sustain a balance between a global and a local mutation operator achieve better performance in a changing leadership strategy than a constant leadership strategy. The results demonstrate that the model is imparted with sufficient dynamics to allow different types of outputs to occur. The artificial worlds approach makes it possible to conduct some experiments that are infeasible to perform in the real world. Combining more selected features into the model would show its potential use in investigating complex human resource management issues.     

动态调整惯性权重的粒子群算法优化

为了优化目前粒子群算法比较容易陷入局部最优、后期收敛过慢等的缺陷,在本文提出了一种改进惯性权重参数来优化算法的方法.其中结合了差分进化算法中的变异算子的操作来提升算法的自适应并且对算法的速度和搜索空间进行边界限制以防止粒子跳出所规定的搜索空间.选择相应的测试函数,使用Matlab软件将提出的改进算法与其他两种算法进行仿真实验对比,结果表明,本文所提出的算法在后期收敛速度以及取得适应度值的稳定性上有一定的提升.     

基于父个体相似度的自适应遗传算法

标准遗传算法在产生后代个体时采用先交叉后变异的策略,一方面当父个体非常相似时,交叉操作很难产生新的个体,影响算法对新的解空间进行搜索,从而导致种群多样性的丧失;另一方面交叉产生的优秀个体再历经变异,极有可能遭破坏而影响算法的收敛性。该文根据染色体的相似性,给出了个体相似度的概念,并在此基础上提出了依据父个体相似度的大小自适应地选择遗传算子(交叉或变异)的遗传算法。仿真实验表明,与采用常规遗传策略的遗传算法相比,新算法能显著提高解的质量和收敛速度。     

基于迁移学习灰支持向量回归机的交互式进化计算

针对机器感知评价和种群进化,提出基于迁移学习灰支持向量回归机的个体适应值预测方法和聚类进化策略.通过共享用户已评价个体适应值学习模型与部分未评价个体适应值学习模型,实现知识模型差异最小化.建立具有迁移学习能力的灰支持向量回归机模型,预测未评价个体适应值.基于聚类子集计算个体平均距离,并设计选择算子和交叉算子,扩大子代搜索区域,增强种群多样性.基于上述策略,采用NSGA-II范式实现交互式进化计算.最后,分析算法时间复杂度,表明算法可提高评价精度,并克服局部收敛问题.将该算法应用于室内灯光调色问题,验证所提出方法的有效性.     

A directional mutation operator for differential evolution algorithms

Differential evolution (DE) is widely studied in the past decade. In its mutation operator, the random variations are derived from the difference of two randomly selected different individuals. Difference vector plays an important role in evolution. It is observed that the best fitness found so far by DE cannot be improved in every generation. In this article, a directional mutation operator is proposed. It attempts to recognize good variation directions and increase the number of generations having fitness improvement. The idea is to construct a pool of difference vectors calculated when fitness is improved at a generation. The difference vector pool will guide the mutation search in the next generation once only. The directional mutation operator can be applied into any DE mutation strategy. The purpose is to speed up the convergence of DE and improve its performance. The proposed method is evaluated experimentally on CEC 2005 test set with dimension 30 and on CEC 2008 test set with dimensions 100 and 1000. It is demonstrated that the proposed method can result in a larger number of generations having fitness improvement than classic DE. It is combined with eleven DE algorithms as examples of how to combine with other algorithms. After its incorporation, the performance of most of these DE algorithms is significantly improved. Moreover, simulation results show that the directional mutation operator is helpful for balancing the exploration and exploitation capacity of the tested DE algorithms. Furthermore, the directional mutation operator modifications can save computational time compared to the original algorithms. The proposed approach is compared with the proximity based mutation operator as both are claimed to be applicable to any DE mutation strategy. The directional mutation operator is shown to be better than the proximity based mutation operator on the five variants in the DE family. Finally, the applications of two real world engineering optimization problems verify the usefulness of the proposed method.     

可用于演化硬件的改进自适应遗传算法研究

为提高演化硬件在演化过程中的收敛速度,以解决其可扩展性问题,研究了标准遗传算法的3个遗传算子,分析了进化不同阶段对遗传算子的不同要求及其对收敛速度的影响.在Srinivas的自适应策略和基于阶段进化的自适应策略的基础上,提出一种新的针对变异算子的自适应策略,并在轮盘赌选择方式中加入适应值标度变换.结合实例,对改进后的算法进行了仿真,结果表明了加入适应值尺度变换和新的自适应策略后,算法的收敛性有所提高.     

Analysis of convergence of an evolutionary algorithm with self-adaptation using a stochastic Lyapunov function

This paper analyses the convergence of evolutionary algorithms using a technique which is based on a stochastic Lyapunov function and developed within the martingale theory. This technique is used to investigate the convergence of a simple evolutionary algorithm with self-adaptation, which contains two types of parameters: fitness parameters, belonging to the domain of the objective function; and control parameters, responsible for the variation of fitness parameters. Although both parameters mutate randomly and independently, they converge to the "optimum" due to the direct (for fitness parameters) and indirect (for control parameters) selection. We show that the convergence velocity of the evolutionary algorithm with self-adaptation is asymptotically exponential, similar to the velocity of the optimal deterministic algorithm on the class of unimodal functions. Although some martingale inequalities have not be proved analytically, they have been numerically validated with 0.999 confidence using Monte-Carlo simulations.     

A new mutation rule for evolutionary programming motivated frombackpropagation learning

Evolutionary programming is mainly characterized by two factors: the selection strategy and the mutation rule. This letter presents a new mutation rule that has the same form as the well-known backpropagation learning rule for neural networks. The proposed mutation rule assigns the best individual's fitness as the temporary target at each generation. The temporal error, the distance between the target and an individual at hand, is used to improve the exploration of the search space by guiding the direction of evolution. The momentum, i.e., the accumulated evolution information for the individual, speeds up convergence. The efficiency and robustness of the proposed algorithm are assessed on several benchmark test functions     

基于线性变换的变异算子及在机器人进化计算中的应用

进化算子对进化计算行为和性能有直接影响.根据进化阶段的不同和个体适应度的大小,采用线性变换法建立能保持群体多样性和收敛性的自适应的变异算子的表达式,并在Evorobot系统中仿真实现.仿真结果表明,基于线性变换的自适应的变异算子改进了进化计算性能,体现了优越性.     

基于学习算子的自学习遗传算法设计

遗传算子是影响遗传算法优化效果的重要因素。针对目前遗传算法研究中忽视个体能动性,没有充分利用进化经验信息的不足,提出反映个体学习能力的学习算子。给出了以个体适应度的变化方向和速度为依据的学习算子设计方法及其计算过程。在此基础上与现有的改进遗传算子结合,提出一种新的改进遗传算法-自学习遗传算法,分析了自学习遗传算法与自适应遗传算法之间在原理上的区别。以一个弹道导弹射程优化问题为算例对算法进行了性能测试,结果表明,在采用相同的改进遗传算子的条件下,学习算子能够以较低的代价提高遗传算法的收敛速度,并获得更好的最终优化结果。     

一种自调整的进化规划算法

针对进化规划算法收敛速度慢和早熟收敛的缺点,将改进的随机搜索方法和进化规划算法相结合,提出了一种自调整的进化规划算法。在该算法中,使用高斯变异算子对个体进行变异,利用改进的随机搜索方法对个体变异进行自调整,提升了个体向适应度高的方向进化的能力,提高了个体间的多样性差异,从而改善算法的性能。对该算法性能进行典型算例的数字仿真证明该算法具有良好的性能。