Swarm and Evolutionary Computation最新文献

In many real-world engineering optimizations, a large number of objective and constraint function values often need to be obtained through simulation software or physical experiments, which incurs significant computational costs and/or time expenses. These problems are known as expensive constraint multi-objective optimization problems (ECMOPs). This paper combines the push and pull search (PPS) framework and proposes a surrogate-assisted evolutionary algorithm to solve ECMOPs through Bayesian active learning, naming it the surrogate-assisted PPS (SA-PPS). Specifically, during the push search stage, candidate solutions are selected based on two indicators: hypervolume improvement and objective uncertainty. These aim to quickly guide the population towards the unconstrained Pareto front while ensuring diversity. During the pull search stage, the population is partitioned into many subregions through reference vectors, and different selection strategies are assigned to each subregion based on its state, aiming to guide the population towards the constrained Pareto front while ensuring diversity. Furthermore, we introduce a batch data selection strategy that utilizes Bayesian active learning to enable the surrogate model to focus on regions of interest in the pull search stage. Extensive experimental results have shown that the proposed SA-PPS algorithm exhibits superior convergence and diversity compared to 9 state-of-the-art algorithms across a variety of benchmark problems and a real-world optimization problem.

With the increasing level of agricultural automation, the combination of agriculture and intelligent vehicle technology is propelling the development of smart agriculture. Although this technology has already been applied widely for various agricultural production tasks, inefficient vehicle scheduling still hasn't been resolved satisfactorily. Oriented towards agricultural harvesting scenarios, a hybrid loading situation vehicle routing problem (HLSVRP) model is proposed to minimize total energy consumption and maximum completion time. A reconstructed multi-objective evolutionary algorithm based on decomposition (R-MOEA/D) is developed to solve the problem. Eight solution representations tailored specifically to the problem are introduced by R-MOEA/D, allowing an extensive exploration of the solution space. A modified Clarke & Wright (MCW) heuristic is designed to generate a high-quality initial population. A novel problem-specific parallel population updating mechanism based on the four crossover and two mutation combinations is also provided to improve the exploration ability. A collaborative search strategy is employed to facilitate cooperation among parallel populations. Finally, a series of comparative experiments conducted on various task scales and vehicle scales verify the effectiveness of the proposed algorithmic components and the exceptional performance for solving HLSVRP.

To alleviate the premature, many evolutionary computation algorithms try to balance the exploitation and exploration by controlling the population diversity. However, randomly diversifying a population cannot always guarantee that an algorithm exploits or explores promising regions. To address this issue, a general framework is proposed in this paper for learning promising regions that are made up of subspaces to guide where to exploit and explore by two reinforcement learning systems. The learning mechanism is as follows: (1) To enhance the efficiency of exploitation, an exploitative reinforcement learning system is constructed to estimate the exploitative potential values of subspaces. Accordingly, basins of attraction are approximated by clustering subspaces and historical solutions are selected within the same basin of attraction to generate new solutions. (2) To efficiently explore the solution space, an explorative reinforcement learning system is established to estimate the explorative potential values of subspaces. Accordingly, algorithms are guided to explore subspaces with higher explorative potential values, promoting the discovery of unexploited promising basins of attraction. The framework is implemented into three conventional evolutionary algorithms, and the mechanism and effectiveness of the implemented algorithms are investigated by comprehensive experimental studies. The experimental results show that the proposed algorithms have competitive performances over the other twelve popular evolutionary algorithms.

This study addresses a hybrid flow shop scheduling problem by considering consistent sublots (HFSP_CS) in lot-streaming. The objective is to minimize the maximum completion time (makespan). By mathematically formulating the HFSP_CS, a mathematical model is established. Next, novel combinations of four meta-heuristics and Q-learning-based improvement tactics are proposed for tackling the related problems for the first time. Drawing upon problem-specific characteristics, five local search operators are employed and selected appropriately by utilizing Q-learning throughout the iterations. Furthermore, the model's veracity is demonstrated through the utilization of the CPLEX solver. Then, by resolving 128 instances, the enhanced algorithms showcase their effectiveness. The results show that the artificial bee colony algorithm integrated with Q-learning is the most competitive algorithm among the tested algorithms.

Addressing the challenges of constrained multiobjective optimization problems (CMOPs) with evolutionary algorithms requires balancing constraint satisfaction and optimization objectives. Coevolutionary multitasking (CEMT) offers a promising strategy by leveraging synergies from distinct, complementary tasks. The primary challenge in CEMT frameworks is constructing suitable auxiliary tasks that effectively complement the main CMOP task. In this paper, we propose an adaptive CEMT framework (ACEMT), which customizes two adaptive auxiliary tasks to enhance CMOP-solving efficiency. The first auxiliary task dynamically narrows constraint boundaries, facilitating exploration in regions with smaller feasible spaces. The second task focuses specifically on individual constraints, continuously adapting to expedite convergence and uncover optimal regions. In solving the main CMOP task, this dual-auxiliary-task strategy not only improves search thoroughness but also clarifies the balance between constraints and objectives. Concretely, ACEMT incorporates an adaptive constraint relaxation technique for the first auxiliary task and a specialized constraint selection strategy for the second. These innovations foster effective knowledge transfer and task synergy, addressing the key challenge of auxiliary task construction in CEMT frameworks. Extensive experiments on three benchmark suites and real-world applications demonstrate ACEMT’s superior performance compared to state-of-the-art constrained evolutionary algorithms. ACEMT sets a new standard in CMOP-solving by strategically constructing and adapting auxiliary tasks, representing a significant advancement in this research direction.

Feature selection (FS) is a crucial technique in machine learning and data mining, serving a variety of purposes such as simplifying model construction, facilitating knowledge discovery, improving computational efficiency, and reducing memory consumption. Despite its importance, the constantly increasing search space of high-dimensional datasets poses significant challenges to FS methods, including issues like the "curse of dimensionality," susceptibility to local optima, and high computational and memory costs. To overcome these challenges, a new FS algorithm named Uniform-solution-driven Binary Feature Selection (UniBFS) has been developed in this study. UniBFS exploits the inherent characteristic of binary algorithms-binary coding-to search the entire problem space for identifying relevant features while avoiding irrelevant ones. To improve the effectiveness and efficiency of the UniBFS algorithm, Redundant Features Elimination algorithm (RFE) is presented in this paper. RFE performs a local search in a very small subspace of the solutions obtained by UniBFS in different stages, and removes the redundant features which do not increase the classification accuracy. Moreover, the study proposes a hybrid algorithm that combines UniBFS with two filter-based FS methods, ReliefF and Fisher, to identify pertinent features during the global search phase. The proposed algorithms are evaluated on 30 high-dimensional datasets ranging from 2000 to 54676 dimensions, and their effectiveness and efficiency are compared with state-of-the-art techniques, demonstrating their superiority.

Internet of Things (IoT) networks consist of physical devices connected to the Internet, embedded with actuators, sensors, and communication components that exchange data. To enhance IoT security, accurately identifying and assessing the safety of connected devices is essential. To improve IoT security, this research proposes the IoT Device Type Identification utilizing Memristor-based Magnetized Hopfield Neural Network with Tuna Swarm Optimization Algorithm (IOT-DTI-MHNN-TSOA). It includes data collection, feature extraction, IoT device type identification. In data collection, an actual network traffic dataset amassed through 10 various IoT device categories is used. In the feature extraction phase, optimal features such as TCP packets' time-to-live by server, packets' inter-arrival time by client, packets' inter-arrival time by server, TCP packets' time-to-live by client, packets' inter-arrival time, packet size, number of bytes sent and received, packet size by client, and total number of packets are extracted using a 2-Dimensional Flexible Analytic Wavelet Transform (2D-FAWT). These extracting features are provided to the IoT device type identification phase. In this phase, a Memristor-based Magnetized Hopfield Neural Network (MHNN) method is employed to perceive the categories of IoT device as known/seen or unknown/unseen categories. The Tuna Swarm Optimization Algorithm (TSOA) enhances the weight parameters of MHNN. The efficacy of the IOT-DTI-MHNN-TSOA classification framework is assessed using performance metrics, like precision, accuracy, F1-score, sensitivity, specificity, error rate, computational time, ROC, Computational Complexity. The IOT-DTI-MHNN-TSOA method provides higher accuracy of 99.97 %, higher sensitivity of 99.95 %, and higher precision of 99.92 % compared to the existing models.

Multimodal multiobjective optimization problems are widely prevalent in real life. Addressing these challenges is crucial as they directly impact the efficiency and effectiveness of solutions across various domains. This paper proposes a novel Multi-Modal Multi-Objective Arithmetic Optimization Algorithm (MMOP-AOA), aimed at achieving a high balance between diversity and convergence in both decision and objective spaces. Arithmetic Optimization Algorithm (AOA) is a highly competitive metaheuristic optimization algorithm with strong exploration and exploitation capabilities. MMOP-AOA extends the AOA for the first time to solve multimodal multiobjective problems, with the following ideas: Firstly, a new exploration and exploitation strategy (NBCNEE) is designed based on the characteristics of AOA.The strategy utilizes Neighborhood-Based Clustering (NBC) to partition the decision space into multiple clusters, aiding MMOP-AOA in capturing more equivalent Pareto subsets (ePSs). Secondly, a convergence and diversity balance mechanism (CDBM) is developed. This mechanism involves comparing the convergence indicator and diversity indicator to select different mutation strategies. Thirdly, an improved crowding distance (ICD) is proposed to address the deficiencies of existing special crowding distance measures. The effectiveness of CDBM and ICD is demonstrated in the paper through experiments on 22 benchmark functions from CEC-2019 and a real-world problem of signal timing optimization at road intersections. The research also reveals that compared to four other advanced multimodal multiobjective optimization algorithms, MMOP-AOA exhibits superior search capability and stability. Furthermore, MMOP-AOA utilizes Neighborhood-Based Clustering (NBC) to partition the decision space into multiple clusters, aiding MMOP-AOA in capturing more equivalent Pareto subsets (ePSs) and provides a theoretical framework for other metaheuristic optimization algorithms to tackle multimodal multiobjective problems.

In this paper, we solve the feature subset selection (FSS) problem with three objective functions namely, cardinality, area under receiver operating characteristic curve (AUC) and Matthews correlation coefficient (MCC) using novel multi-objective evolutionary algorithms (MOEAs). MOEAs often encounter poor convergence due to the increase in non-dominated solutions and getting entrapped in the local optima. This situation worsens when dealing with large, voluminous big and high-dimensional datasets. To address these challenges, we propose parallel, fractional dominance-based MOEAs for FSS under Spark. Further, to improve the exploitation of MOEAs, we introduce a novel batch opposition-based learning (BOP) along with a cardinality constraint on the opposite solution. Accordingly, we propose two variants, namely, BOP1 and BOP2. In BOP1, a single neighbour is randomly chosen in the opposite solution space, whereas in BOP2, a group of randomly chosen neighbours in the opposite solution space. In either case, the opposite solutions are evaluated to improve the exploitation capability of the underlying MOEAs. We observe that in terms of mean optimal objective function values and across all datasets, the proposed BOP2 variant of parallel fractional dominance-based algorithms emerges as the top performer in obtaining efficient solutions. Further, we introduce a novel metric, namely the ratio of hypervolume (HV) and inverted generated distance (IGD), HV/IGD, that combines both diversity and convergence. With respect to the mean HV/IGD computed over 20 runs and Formula 1 racing, the BOP1 variants of fractional dominance-based MOEAs outperformed other algorithms.

This paper establishes the mathematical model for Storage Location Assignment (SLA) problem in large-scale automated warehouses by combining three objectives: efficiency, shelf stability, and stacker load balancing. Along with a novel repair strategy to handle the complex constraints of large-scale problems. Additionally, a coding method and solution approach suitable for practical application scenarios are developed. In order to solve large-scale SLA problem, an improved vortex search algorithm is proposed based on attraction operation in flow field, dimension-by-dimension dynamic radius and leadership decision-making mechanism (FDVSA). In the experimental part, the algorithm effectiveness experiment of FDVSA was first conducted using the large-scale global optimization test sets IEEE congress on evolutionary computation 2010 and 2013 (CEC2010, CEC2013). The results show that: (1) Compared with other comparison algorithms, the comprehensive average optimization rate of FDVSA in CEC2010 and CEC2013 is 88 % and 78 %, respectively. (2) The experimental results of FDVSA showed that each improvement strategy has advantages in dealing with large-scale problems. (3) The post-hoc analysis showed that there are significant differences between FDVSA and other comparison algorithms, and FDVSA is significantly better. Finally, FDVSA and other comparison algorithms are solved on three different scale and complexity of SLA cases. The results show that: (1) FDVSA has significant advantages in solving large-scale SLA problem, and the comprehensive average optimization rate is 19 %. (2) The convergence curve and boxplot showed that FDVSA has good convergence speed and solving stability. (3) The effectiveness of the repair strategy was verified by experiments in the large-scale SLA problems.