Convergence of multi-objective evolutionary algorithms to a uniformly distributed representation of the Pareto front

In evolutionary multi-objective optimization (EMO), the convergence to the Pareto set of a multi-objective optimization problem (MOP) and the diversity of the final approximation of the Pareto front are two important issues. In the existing definitions and analyses of convergence in multi-objective evolutionary algorithms (MOEAs), convergence with probability is easily obtained because diversity is not considered. However, diversity cannot be guaranteed. By combining the convergence with diversity, this paper presents a new definition for the finite representation of a Pareto set, the B-Pareto set, and a convergence metric for MOEAs. Based on a new archive-updating strategy, the convergence of one such MOEA to the B-Pareto sets of MOPs is proved. Numerical results show that the obtained B-Pareto front is uniformly distributed along the Pareto front when, according to the new definition of convergence, the algorithm is convergent.     

均衡分布性与收敛性的协同进化多目标优化算法

为了进一步提升多目标进化算法(MOEAs)的收敛速度和解集分布性,针对变量无关问题,借助合作型协同进化模型,提出一种均衡分布性与收敛性的协同进化多目标优化算法(CMOA-BDC). CMOA-BDC 首先设置一个精英集合,采用支配关系从进化种群与精英集合中选择首层,并用拥挤距离保持其分布性;然后运用聚类将首层分类,并建立相应概率模型;最后通过模拟退火组合分布估计与遗传进化,达到协同进化.通过与经典 MOEAs 比较的结果表明, CMOA-BDC 获得的解集具有更好的收敛性和分布性.     

基于改进K 支配排序的高维多目标进化算法

为提高4目标以上高维多目标优化问题的求解性能,提出一种基于改进K支配排序的高维多目标进化算法(KS-MODE).该算法针对K支配的支配关系和排序方法进行改进,避免循环支配并增强选择压力;设计新的全局密度估计方法提高局部密度估计精确性;设计新的精英选择策略和适应度值评价函数;采用CAO局部搜索算子加速收敛.在4～30个目标标准测试函数上的实验结果表明,KS-MODE能够在保证解集分布性的同时大幅提升收敛性和稳定性,能够有效求解高维多目标优化问题.     

Theory and Software Implementations of Shor’s <Emphasis Type="Italic">r</Emphasis>-Algorithms<Superscript>*</Superscript>

Three computational forms of r-algorithms with different amount of computation per iteration are considered. The results on the convergence of the limit variant of r-algorithms for convex smooth functions and the r _μ (α)-algorithm for convex piecewise smooth functions are presented. Practical aspects of the variant of r (α)-algorithms with a constant coefficient of space dilation α and an adaptive method for step adjustment in the direction of the normalized anti-subgradient in the transformed space of variables are discussed.     

一种基于群体分布特征的自适应多目标粒子群优化算法

针对多目标优化问题求解,提出基于群体分布特征的多目标自适应粒子群优化算法(pdMOPSO).首先借助统计方法分析归档集在决策空间的分布特征,以此划分进化状态,指导全局引导粒子的选择;然后设计粒子重排策略,动态调控种群的分布;最后依据进化状态设计不同的归档集维护策略,实现归档集中分布性和收敛性的均衡.以ZDT、DTLZ和CEC09为测试集,与7种多目标优化算法对比,指标IGD、Spread和ER结果表明,所提出的算法在收敛性和分布性上均有显著优势.     

A multi-objective artificial bee colony algorithm based on division of the searching space

This paper presents a new multi-objective artificial bee colony algorithm called dMOABC by dividing the whole searching space S into two independent parts S ₁ and S ₂. In this algorithm, two ”basic” colonies are assigned to search potential solutions in regions S ₁ and S ₂, while the so-called ”synthetic” colony explores in S. This multi-colony model could enable the good diversity of the population, and three colonies share information in a special way. A fixed-size external archive is used to store the non-dominated solutions found so far. The diversity over the archived solutions is controlled by utilizing a self-adaptive grid. For basic colonies, neighbor information is used to generate new food sources. For the synthetic colony, besides neighbor information, the global best food source gbest selected from the archive, is also adopted to guide the flying trajectory of both employed and onlooker bees. The scout bees are used to get rid of food sources with poor qualities. The proposed algorithm is evaluated on a set of unconstrained multi-objective test problems taken from CEC09, and is compared with 11 other state-of-the-art multi-objective algorithms by applying Friedman test in terms of four indicators: HV, SPREAD, EPSILON and IGD. It is shown by the test results that our algorithm significantly surpasses its competitors.     

Two Plane-Probing Algorithms for the Computation of the Normal Vector to a Digital Plane

Digital planes are sets of integer points located between two parallel planes. We present a new algorithm that computes the normal vector of a digital plane given only a predicate “is a point x in the digital plane or not”. In opposition to classical recognition algorithm, this algorithm decides on-the-fly which points to test in order to output at the end the exact surface characteristics of the plane. We present two variants: the H-algorithm, which is purely local, and the R-algorithm which probes further along rays coming out from the local neighborhood tested by the H-algorithm. Both algorithms are shown to output the correct normal to the digital planes if the starting point is a lower leaning point. The worst-case time complexity is in \(O(\omega )\) for the H-algorithm and \(O(\omega \log \omega )\) for the R-algorithm, where \(\omega \) is the arithmetic thickness of the digital plane. In practice, the H-algorithm often outputs a reduced basis of the digital plane while the R-algorithm always returns a reduced basis. Both variants perform much better than the theoretical bound, with an average behavior close to \(O(\log \omega )\). Finally, we show how this algorithm can be used to analyze the geometry of arbitrary digital surfaces, by computing normals and identifying convex, concave or saddle parts of the surface. This paper is an extension of Lachaud et al. (Proceedings of 19th IAPR international conference discrete geometry for computer imagery (DGCI’2016), Nantes, France. Springer, Cham, 2016).     

自适应偏好半径划分区域的多目标进化方法

偏好多目标进化算法是一类帮助决策者找到感兴趣的Pareto最优解的算法.目前,在以参考点位置作为偏好信息载体的偏好多目标进化算法中,不合适的参考点位置往往会严重影响算法的收敛性能,偏好区域的大小难以控制,在高维问题上效果较差.针对以上问题,通过计算基于种群的自适应偏好半径,利用自适应偏好半径构造一种新的偏好关系模型,通过对偏好区域进行划分,提出基于偏好区域划分的偏好多目标进化算法.将所提算法与4种常用的以参考点为偏好信息载体的多目标进化算法g-NSGA-II、r-NSGA-II、角度偏好算法、MOEA/D-PRE进行对比实验,结果表明,所提算法具有较好的收敛性能和分布性能,决策者可以控制偏好区域大小,在高维问题上也具有较好的收敛效果.     

Multiobjective Markov chains optimization problem with strong Pareto frontier: Principles of decision making

In this paper, we present a novel approach for computing the Pareto frontier in Multi-Objective Markov Chains Problems (MOMCPs) that integrates a regularized penalty method for poly-linear functions. In addition, we present a method that make the Pareto frontier more useful as decision support system: it selects the ideal multi-objective option given certain bounds. We restrict our problem to a class of finite, ergodic and controllable Markov chains. The regularized penalty approach is based on the Tikhonov’s regularization method and it employs a projection-gradient approach to find the strong Pareto policies along the Pareto frontier. Different from previous regularized methods, where the regularizator parameter needs to be large enough and modify (some times significantly) the initial functional, our approach balanced the value of the functional using a penalization term (μ) and the regularizator parameter (δ) at the same time improving the computation of the strong Pareto policies. The idea is to optimize the parameters μ and δ such that the functional conserves the original shape. We set the initial value and then decrease it until each policy approximate to the strong Pareto policy. In this sense, we define exactly how the parameters μ and δ tend to zero and we prove the convergence of the gradient regularized penalty algorithm. On the other hand, our policy-gradient multi-objective algorithms exploit a gradient-based approach so that the corresponding image in the objective space gets a Pareto frontier of just strong Pareto policies. We experimentally validate the method presenting a numerical example of a real alternative solution of the vehicle routing planning problem to increase security in transportation of cash and valuables. The decision-making process explored in this work correspond to the most frequent computational intelligent models applied in practice within the Artificial Intelligence research area.     

Convergence performance comparison of quantum-inspired multi-objective evolutionary algorithms

In recent research, we proposed a general framework of quantum-inspired multi-objective evolutionary algorithms (QMOEA) and gave one of its sufficient convergence conditions to the Pareto optimal set. In this paper, two Q-gate operators, H gate and R&N gate, are experimentally validated as two Q-gate paradigms meeting the convergence condition. The former is a modified rotation gate, and the latter is a combination of rotation gate and NOT gate with the specified probability. To investigate their effectiveness and applicability, several experiments on the multi-objective 0/1 knapsack problems are carried out. Compared to two typical evolutionary algorithms and the QMOEA only with rotation gate, the QMOEA with H gate and R&N gate have more powerful convergence ability in high complex instances. Moreover, the QMOEA with R&N gate has the best convergence in almost all of the experimental problems. Furthermore, the appropriate ε value regions for two Q-gates are verified.     

A self-adaptive multi-objective harmony search algorithm based on harmony memory variance

Although harmony search (HS) algorithm has shown many advantages in solving global optimization problems, its parameters need to be set by users according to experience and problem characteristics. This causes great difficulties for novice users. In order to overcome this difficulty, a self-adaptive multi-objective harmony search (SAMOHS) algorithm based on harmony memory variance is proposed in this paper. In the SAMOHS algorithm, a modified self-adaptive bandwidth is employed, moreover, the self-adaptive parameter setting based on variation of harmony memory variance is proposed for harmony memory considering rate (HMCR) and pitch adjusting rate (PAR). To solve multi-objective optimization problems (MOPs), the proposed SAMOHS uses non-dominated sorting and truncating procedure to update harmony memory (HM). To demonstrate the effectiveness of the SAMOHS, it is tested with many benchmark problems and applied to solve a practical engineering optimization problem. The experimental results show that the SAMOHS is competitive in convergence performance and diversity performance, compared with other multi-objective evolutionary algorithms (MOEAs). In the experiment, the impact of harmony memory size (HMS) on the performance of SAMOHS is also analyzed.     

Multi-objective optimal reactive power dispatch considering voltage stability in power systems using HFMOEA

This paper proposes a new hybrid fuzzy multi-objective evolutionary algorithm (HFMOEA) based approach for solving complex multi-objective, mixed integer nonlinear problems such as optimal reactive power dispatch considering voltage stability (ORPD-VS). In HFMOEA based optimization approach, the two parameters like crossover probability (P_C) and mutation probability (P_M) are varied dynamically through the output of a fuzzy logic controller. The fuzzy logic controller is designed on the basis of expert knowledge to enhance the overall stochastic search capability for generating better pareto-optimal solution. Two detailed case studies are presented: Firstly, the performance of HFMOEA is tested on five benchmark test problems such as ZDT1, ZDT2, ZDT3, ZDT4 and ZDT6 as suggested by Zitzler, Deb and Thiele; Secondly, HFMOEA is applied to multi-objective ORPD-VS problem. In both the case studies, the optimization results obtained from HFMOEA are analysed and compared with the same obtained from two versions of elitist non-dominated sorting genetic algorithms such as NSGA-II and MNSGA-II in terms of various performance metrics. The simulation results are promising and confirm the ability of HFMOEA for generating better pareto-optimal fronts with superior convergence and diversity.     

自适应小生境遗传算法在系统级综合中的应用

多目标遗传算法是解决SoC系统级综合问题的有效途径之一,但现有的遗传算法只能求得非劣解集前沿的一部分,局部搜索能力差,收敛速度较慢。该文通过结合小生境技术,根据种群往代的多样性信息,自适应地确定子种群的规模和交叉、变异的概率,提出一种自适应小生境遗传算法,有效提高解集的覆盖率,加快收敛速度。以视频编解码的系统级综合为例,证明该算法可以较快地产生较多非 劣解。     

Clear and smooth path planning

One of the challenging problems in motion planning is finding an efficient path for a robot in different aspects such as length, clearance and smoothness. We formulate this problem as two multi-objective path planning models with the focus on robot's energy consumption and path's safety. These models address two five- and three-objectives optimization problems. We propose an evolutionary algorithm for solving the problems. For efficient searching and achieving Pareto-optimal regions, in addition to the standard genetic operators, a family of path refiner operators is introduced. The new operators play a local search role and intensify power of the algorithm in both explorative and exploitative terms. Finally, we verify the models and compare efficiency of the algorithm and the refiner operators by other multi-objective algorithms such as strength Pareto evolutionary algorithm 2 and multi-objective particle swarm optimization on several complicated path planning test problems.     

Discrete-time dynamic average consensus

We propose a class of discrete-time dynamic average consensus algorithms that allow a group of agents to track the average of their reference inputs. The convergence results rely on the input-to-output stability properties of static average consensus algorithms and require that the union of communication graphs over a bounded period of time be strongly connected. The only requirement on the set of reference inputs is that the maximum relative deviation between the nth-order differences of any two reference inputs be bounded for some integer n≥1.     

Accelerating the quadratic lower-bound algorithm via optimizing the shrinkage parameter

When the Newton-Raphson algorithm or the Fisher scoring algorithm does not work and the EM-type algorithms are not available, the quadratic lower-bound (QLB) algorithm may be a useful optimization tool. However, like all EM-type algorithms, the QLB algorithm may also suffer from slow convergence which can be viewed as the cost for having the ascent property. This paper proposes a novel ‘shrinkage parameter’ approach to accelerate the QLB algorithm while maintaining its simplicity and stability (i.e., monotonic increase in log-likelihood). The strategy is first to construct a class of quadratic surrogate functions Q_r(θ|θ^(t)) that induces a class of QLB algorithms indexed by a ‘shrinkage parameter’ r (r∈R) and then to optimize r over R under some criterion of convergence. For three commonly used criteria (i.e., the smallest eigenvalue, the trace and the determinant), we derive a uniformly optimal shrinkage parameter and find an optimal QLB algorithm. Some theoretical justifications are also presented. Next, we generalize the optimal QLB algorithm to problems with penalizing function and then investigate the associated properties of convergence. The optimal QLB algorithm is applied to fit a logistic regression model and a Cox proportional hazards model. Two real datasets are analyzed to illustrate the proposed methods.     

A hybridized angle-encouragement-based decomposition approach for many-objective optimization problems

Due to the large objective space when handling many-objective optimization problems (MaOPs), it is a challenging work for multi-objective evolutionary algorithms (MOEAs) to balance convergence and diversity during the search process. Although a number of decomposition-based MOEAs have been designed for the above purpose, some difficulties are still encountered for tackling some difficult MaOPs. As inspired by the existing decomposition approaches, a new Hybridized Angle-Encouragement-based (HAE) decomposition approach is proposed in this paper, which is embedded into a general framework of decomposition-based MOEAs, called MOEA/D-HAE. Two classes of decomposition approaches, i.e., the angle-based decomposition and the proposed encouragement-based boundary intersection decomposition, are sequentially used in HAE. The first one selects appropriate solutions for association in the feasible region of each subproblem, which is expected to well maintain the diversity of associated solutions. The second one acts as a supplement for the angle-based one under the case that no solution is located in the feasible region of subproblem, which aims to ensure the convergence and explore the boundaries. By this way, HAE can effectively combine their advantages, which helps to appropriately balance convergence and diversity in evolutionary search. To study the effectiveness of HAE, two series of well-known test MaOPs (WFG and DTLZ) are used. The experimental results validate the advantages of HAE when compared to other classic decomposition approaches and also confirm the superiority of MOEA/D-HAE over seven recently proposed many-objective evolutionary algorithms.     

Subexponential algorithms for partial cover problems

Partial Cover problems are optimization versions of fundamental and well-studied problems like Vertex Cover and Dominating Set. Here one is interested in covering (or dominating) the maximum number of edges (or vertices) using a given number k of vertices, rather than covering all edges (or vertices). In general graphs, these problems are hard for parameterized complexity classes when parameterized by k. It was recently shown by Amini et al. (2008) [1] that Partial Vertex Cover and Partial Dominating Set are fixed parameter tractable on large classes of sparse graphs, namely H-minor-free graphs, which include planar graphs and graphs of bounded genus. In particular, it was shown that on planar graphs both problems can be solved in time ²O(k)nO(1).During the last decade there has been an extensive study on parameterized subexponential algorithms. In particular, it was shown that the classical Vertex Cover and Dominating Set problems can be solved in subexponential time on H-minor-free graphs. The techniques developed to obtain subexponential algorithms for classical problems do not apply to partial cover problems. It was left as an open problem by Amini et al. (2008) [1] whether there is a subexponential algorithm for Partial Vertex Cover and Partial Dominating Set. In this paper, we answer the question affirmatively by solving both problems in time not only on planar graphs but also on much larger classes of graphs, namely, apex-minor-free graphs. Compared to previously known algorithms for these problems our algorithms are significantly faster and simpler.     

一种新型的多目标优化混合量子进化算法

针对复杂多目标优化问题,提出一种混合量子进化算法,并利用它求解多目标函数优化问题。该算法根据多目标优化的特点,创建外部集合保存历代搜索到的非支配解,利用其中的精英个体设计了一种旋转角自适应调整的量子门更新策略,并对量子比特表示的概率幅设置最大和最小阈值,以防止量子群体早熟收敛。借鉴量子门引入了专门针对量子个体的旋转交叉算子,同时小概率地对量子比特进行取反变异操作。对所提算法的计算复杂度进行了理论分析。与另一种已有的多目标量子进化算法的比较结果表明,所提算法具有更好的收敛性能、分布特性及求解效率。     

A novel strategy of pareto-optimal solution searching in multi-objective particle swarm optimization (MOPSO)

In multi-objective particle swarm optimization (MOPSO) algorithms, finding the global optimal particle (gBest) for each particle of the swarm from a set of non-dominated solutions is very difficult yet an important problem for attaining convergence and diversity of solutions. First, a new Pareto-optimal solution searching algorithm for finding the gBest in MOPSO is introduced in this paper, which can compromise global and local searching based on the process of evolution. The algorithm is implemented and is compared with another algorithm which uses the Sigma method for finding gBest on a set of well-designed test functions. Finally, the multi-objective optimal regulation of cascade reservoirs is successfully solved by the proposed algorithm.