Chenyang Bu

Continuous Dynamic Constrained Optimization With Ensemble of Locating and Tracking Feasible Regions Strategies

Dynamic constrained optimization problems (DCOPs) are difficult to solve because both objective function and constraints can vary with time. Although DCOPs have drawn attention in recent years, little work has been performed to solve DCOPs with multiple dynamic feasible regions from the perspective of locating and tracking multiple feasible regions in parallel. Moreover, few benchmarks have been proposed to simulate the dynamics of multiple disconnected feasible regions. In this paper, first, the idea of tracking multiple feasible regions, originally proposed by Nguyen and Yao, is enhanced by specifically adopting multiple subpopulations. To this end, the dynamic species-based particle swam optimization (DSPSO), a representative multipopulation algorithm, is adopted. Second, an ensemble of locating and tracking feasible regions strategies is proposed to handle different types of dynamics in constraints. Third, two benchmarks are designed to simulate the DCOPs with dynamic constraints. The first benchmark, including two variants of G24 (called G24v and G24w), could control the size of feasible regions. The second benchmark, named moving feasible regions benchmark (MFRB), is highly configurable. The global optimum of MFRB is calculated mathematically for experimental comparisons. Experimental results on G24, G24v, G24w, and MFRB show that the DSPSO with the ensemble of strategies performs significantly better than the original DSPSO and other typical algorithms.

Species-based Particle Swarm Optimizer enhanced by memory for dynamic optimization

Description of the states in both before updating and after updating.Figure gives an example to illustrate our algorithm. Part (a) gives the state of before updating. Part (b) explains the state of after updating. Assume there are five peaks in the search space, numbered 1-5 in Part (b). However, only four peaks are detected by the current population, numbered 1-4 in Part (a).Where, ź denotes a replacer from the memory, Δ denotes a new generated particle around the replacer, and ź denotes a replaced particle in the population. We can see that no more than one particle is replaced in each sub-population.Since ź in peak 5 is better than the closest seed, i.e. the seed in sub-population 2, and its distance to the closest seed is larger than rs, the sub-population 5 is created to exploit this area.In sub-population 3, ź is better than the closest seed, i.e. the seed in sub-population 3, and the distance between it to the closest seed is larger than 0.5×rs and less than rs, so only one Δ is created around the ź.In sub-population 2, ź is better than the closest seed, i.e. the seed in sub-population 2, but the distance between it to the closest seed is less than 0.5×rs, so only ź is added to this sub-population.There is one case that has not been described in this example. Here we suppose the replacer is not better than the worst particle in sub-population 1, so no replacement is conducted in this sub-population.Display Omitted A new population updating method is proposed to enhance a representative algorithm, i.e. the Species-based Particle Swarm Optimization.Experimental results show that the MSPSO is competitive on MPB, CMPB and DRGDB.The effect of the memory size on the performance of the proposed algorithm is tested. Both the species strategy and the memory scheme are efficient methods for addressing dynamic optimization problems. However, the combination of these two efficient techniques has scarcely been studied. Thus, this paper focuses on how to hybridize these two methods. In this paper, a new swarm updating method is proposed to enhance a representative species-based algorithm, i.e., SPSO (Species-based Particle Swarm Optimization), and the new algorithm is named MSPSO. MSPSO has two characteristics. First, the number of replaced particles in the current swarm is set adaptively according to the number of species. To not substantially destroy the exploitation capability of each species, no more than one particle in each species is replaced by the memory. Second, the retrieved memory particles are categorized according to their fitness values and their distances to the seed of the closest species. Aimed at enhancing the search in both promising areas and existing species, each category is processed by different operations. The MPB, Cyclic MPB and DRPBG are used to test the performance of MSPSO. Experimental results demonstrate that MSPSO is competitive for dynamic optimization problems.

Evolutionary community discovery in dynamic networks based on leader nodes

Evolutionary community discovery is a hot research topic which clusters the dynamic or temporal network. The communities detected in dynamic network should get reasonable partition for the current data while simultaneously not deviate drastically from the previous ones. In this paper, the evolutionary community discovery algorithm based on leader nodes (EvoLeaders) is proposed to cluster the dynamic network. Compared with the static community discovery algorithm based on leader nodes (the Top Leaders algorithm), experimental results over two real-world datasets demonstrate that the EvoLeaders is more suitable for dynamic scenarios.

Differential evolution with a species-based repair strategy for constrained optimization

Evolutionary Algorithms (EAs) with gradient-based repair, which utilize the gradient information of the constraints set, have been proved to be effective. It is known that it would be time-consuming if all infeasible individuals are repaired. Therefore, so far the infeasible individuals to be repaired are randomly selected from the population and the strategy of choosing individuals to be repaired has not been studied yet. In this paper, the Species-based Repair Strategy (SRS) is proposed to select representative infeasible individuals instead of the random selection for gradient-based repair. The proposed SRS strategy has been applied to εDEag which repairs the random selected individuals using the gradient-based repair. The new algorithm is named SRS-εDEag. Experimental results show that SRS-εDEag outperforms εDEag in most benchmarks. Meanwhile, the number of repaired individuals is reduced markedly.

Accelerate Population-Based Stochastic Search Algorithms With Memory for Optima Tracking on Dynamic Power Systems

Existing population-based Stochastic Search Algorithms (SSAs) are too time-consuming to solve dynamic optimal power flow (OPF). The solution proposed in this paper is to accelerate SSAs with memory. Two memory schemes, the similarity retrieval scheme and the mean-based immigrants scheme, are proposed and applied together to the Differential Evolution and Particle Swarm Optimizer, which are two representatives of SSAs. Experiments are conducted on modified IEEE 30-bus and IEEE 118-bus systems with changing load buses and the objective of minimizing real power transmission loss. The results show that the proposed schemes significantly improve the performance of the two existing algorithms, and that SSAs could be practical for tracking optima of dynamic OPF.

Solving online dynamic time-linkage problems under unreliable prediction

Graphical abstractAn example of false optimum. The solid line represents the real function, and the dashed line represents a predicted function based on the collected data (the black dots). Display Omitted HighlightsFirst, a stochastic-ranking-based selection scheme is designed to improve the existing prediction approach under unreliable prediction.Second, the experimental results show that the improved algorithm outperforms the original prediction approach in most of the tested cases. Dynamic time-linkage optimization problems (DTPs) are a special class of dynamic optimization problems (DOPs) with the feature of time-linkage. Time-linkage means that the decisions taken now could influence the problem states in future. Although DTPs are common in practice, attention from the field of evolutionary optimization is little. To date, the prediction method is the major approach to solve DTPs in the field of evolutionary optimization. However, in existing studies, the method of how to deal with the situation where the prediction is unreliable has not been studied yet for the complete Black-Box Optimization (BBO) case. In this paper, the prediction approach EA+predictor, proposed by Bosman, is improved to handle such situation. A stochastic-ranking selection scheme based on the prediction accuracy is designed to improve EA+predictor under unreliable prediction, where the prediction accuracy is based on the rank of the individuals but not the fitness. Experimental results show that, compared with the original prediction approach, the performance of the improved algorithm is competitive.

Clustering spatial data by the neighbors intersection and the density difference

Clustering is a classical unsupervised learning task, which is aimed to divide a data set into several groups with similar objects. Clustering problem has been studied for many years, and many excellent clustering algorithms have been proposed. In this paper, we propose a novel clustering method based on density, which is simple but effective. The primary idea of the proposed method is given as follows. Firstly, the point with the largest local density in a cluster is considered as the cluster center. The local density of each point is estimated based on the distance (called radius) between the point and its k-th nearest neighbor. The point with a smaller radius indicates a larger local density. Secondly, the difference of the local densities between each two internal points should be small, while the difference between the density of a border point and the density of an internal point should be relatively large. Thirdly, if the intersection of k nearest neighbors of two points is small, they should be assigned to different clusters. The proposed algorithm has been compared with a typical clustering algorithm named FDPCluster, and the experimental results show that our algorithm has better clustering quality.

Clustering Evolutionary Data with an r-Dominance Based Multi-objective Evolutionary Algorithm

Clustering evolutionary data (or called evolutionary clustering) has received an enormous amount of attention in recent years. A recent framework (called temporal smoothness) considers that the clustering result should depend mainly on the current data while simultaneously not deviate too much from previous ones. In this paper, evolutionary data is clustered by a multi-objective evolutionary algorithm based on r-dominance, and the corresponding algorithm is named rEvoC. The rEvoC considers the previous clustering result (or historical data) as the reference point. We propose three strategies to define the reference point and to calculate the distance between a reference point and an individual. Based on the reference point and the r-dominance relation, the search could be guided into the region, in which a solution not only could cluster the current data well, but also does not shift two much from the previous one. Additionally, the rEvoC adopts one step k-means as a local search operator to accelerate the evolutionary search. Experimental results on two different data sets are given. The experimental results demonstrate that, the rEvoC achieves better performance than the corresponding static clustering algorithm and the evolutionary k-means algorithm.

A Hybrid Particle Swarm Optimization for High-Dimensional Dynamic Optimization

High-Dimensional Dynamic Optimization Problems (HDDOPs) commonly exist in real-world applications. In evolutionary computation field, most of existing benchmark problems, which could simulate HDDOPs, are non-separable. Thus, we give a novel benchmark problem, called high-dimensional moving peaks benchmark to simulate separable, partially separable, and non-separable problems. Moreover, a hybrid Particle Swarm Optimization algorithm based on Grouping, Clustering and Memory strategies, i.e. GCM-PSO, is proposed to solve HDDOPs. In GCM-PSO, a differential grouping method is used to decompose a HDDOP into a number of sub-problems based on variable interactions firstly. Then each sub-problem is solved by a species-based particle swarm optimization, where the nearest better clustering is adopted as the clustering method. In addition, a memory strategy is also adopted in GCM-PSO. Experimental results show that GCM-PSO performs better than the compared algorithms in most cases.

Improved CFDP Algorithms Based on Shared Nearest Neighbors and Transitive Closure

A recently proposed clustering algorithm named Clustering by fast search and Find of Density Peaks (CFDP) can automatically identify the cluster centers without an iterative process. The key step in CFDP is searching for the nearest neighbor with higher density for each point. However, the CFDP algorithm may not be effective for cases in which there exist density fluctuations within a cluster or between two nearby clusters. In this study, two improved algorithms named CFDP-ED-TSNN1 and CFDP-ED-TSNN2 are presented, which adopt different ways to utilize the dissimilarity. Here, the dissimilarity is based on shared nearest neighbors and transitive closure. The experimental results on both several artificial datasets and a real-world dataset show that the improved algorithms are competitive.