Swarm and Evolutionary Computation最新文献

The automation of meta-heuristic algorithm configuration holds the utmost significance in evolutionary computation. A hierarchical reinforcement learning-aware hyper-heuristic algorithm with fitness landscape analysis (HRLHH) is proposed to flexibly configure the suitable algorithms under various optimization scenarios. Two kinds of fitness landscape analysis techniques improved based on specific problem characteristics construct the state spaces for hierarchical reinforcement learning. Among them, an adaptive classification based on dynamic ruggedness of information entropy is designed to discern the complexity of problems, which serves as the basis for decision-making actions in upper-layer space. Additionally, an online dispersion metric based on knowledge is further presented to distinguish the precise landscape features in lower-layer space. In light of the characteristics of the state spaces, the hierarchical action spaces composed of meta-heuristics with disparate exploration and exploitation are designed, and various action selection strategies are introduced. Taking into account the real-time environment and algorithm evolution behavior, dynamic reward mechanisms based on evolutionary success rate and population convergence rate are utilized to enhance search efficiency. The experimental results on the IEEE Congress on Evolutionary Computation (CEC) 2017, CEC 2014, and large-scale CEC 2013 test suites demonstrate that the proposed HRLHH exhibits superiority in terms of accuracy, stability, and convergence speed, and possesses strong generalization.

The role of multi-area economic dispatch (MAED) in power system operation is increasingly significant. It is a non-linear and multi-constraint problem with many local extremes when considering the valve point effects, posing challenges in obtaining a globally optimal solution, especially for large-scale systems. In this study, an improved variant of differential evolution (DE) called CLDE based on comprehensive learning strategy (CLS) is proposed to solve this problem. Three improved strategies are employed to enhance the performance of CLDE. (1) A CLS-based guided mutation strategy is proposed, in which learning exemplars constructed by competent individuals are used to generate mutant vectors to prevent the searching away from global optimum and speed up convergence. (2) A time-varying increasing crossover rate is devised. It can endow CLDE with a larger probability at the later stage to help individuals escape from local extremes. (3) A CLS-based crossover strategy is presented. Trial vectors directly utilize the information from learning exemplars for evolving, which can ensure the search efficiency and population diversity. CLDE is applied to six MAED cases. Compared with DE, it approximately consumes 32 %, 35 %, 10 %, 22 %, 62 %, and 20 % of evaluations to attain comparable results, saves 126.2544$/h, 81.8173$/h, 152.0660$/h, 360.7907$/h, 65.5757$/h, and 1732.8544$/h in fuel costs on average, and exhibits improvements of 34.77 %, 1.80 %, 0.00 %, 76.09 %, 95.15 %, and 16.76 % in robustness, respectively. Moreover, it also outperforms other state-of-the-art algorithms significantly in statistical analysis. Furthermore, the effects of improved strategies on CLDE are thoroughly investigated.

With the widespread adoption of intelligent transportation equipment such as AGVs in the manufacturing field, the flexible job shop scheduling considering limited transportation resources has increasingly attracted attention. However, current research does not consider various dynamic disturbances in real production scenarios, resulting in lower executability of scheduling solutions. To solve this problem, a dynamic flexible job shop scheduling model with limited transportation resources is established, aiming to minimize makespan and total energy consumption. Considering three types of disturbances: job cancellation, machine breakdown, and AGV breakdown, the corresponding event-driven rescheduling strategy is proposed, and a rescheduling instability index is designed to measure the performance of the rescheduling strategy. A $Q$-Learning-based NSGA-II algorithm (QNSGA-II) is proposed. By learning the feedback historical search experience, it adaptively selects the appropriate neighborhood structures for local search; and a hybrid initialization strategy tailored to the problem characteristics is designed to improve the optimization performance of the algorithm. Through simulation experiments, the effectiveness of the rescheduling strategies and the superiority of the QNSGA-II algorithm in solving such problems are validated.

With the continuous advancement of the manufacturing industry, many random disturbances gradually appear in the job shop production process, such as the insertion of new jobs or random machine failures. This type of disruption often creates production chaos and scheduling problems. This paper takes new job insertion and random machine failure as dynamic events. It establishes an optimization model for dynamic flexible job shop scheduling (DFJSP) problems to cope with this issue. A two-stage double deep Q-network, TS-DDQN algorithm is proposed based on an improved deep reinforcement learning algorithm to solve the complex multi-objective DFJSP optimization problem by adopting two-stage decision-making. In the TS-DDQN, an external non-dominated set is used to enhance the training speed of the network and the quality of the solution, which can store the non-dominated solutions. Moreover, the solutions on the Pareto front are used to train the network parameters. Extensive experimentation is conducted on benchmark datasets to evaluate the performance of the proposed algorithm against the existing scheduling methods. The outcomes underscore the superior efficacy of the proposed algorithm concerning solution quality, convergence speed, and adaptability within dynamic environments. This research contributes to advancing the methods in solving multi-objective DFJSP problems and highlights the potential of deep reinforcement learning to yield managerial advantages in manufacturing industries.

Using infeasible information to balance objective optimization and constraint satisfaction is a very promising research direction to address constrained multi-objective problems (CMOPs) via evolutionary algorithms (EAs). The existing constrained multi-objective evolutionary algorithms (CMOEAs) still face the issue of striking a good balance when solving CMOPs with diverse characteristics. To alleviate this issue, in this paper we develop a two-stage different evolution with a dynamic population assignment strategy for CMOPs. In this approach, two cooperative populations are used to provide feasible driving forces and infeasible guiding knowledge. To adequately utilize the infeasibility information, a dynamic population assignment model is employed to determine the primary population, which is used as the parents to generate offspring. The entire search process is divided into two stages, in which the two populations work in weak and strong cooperative ways, respectively. Furthermore, multistrategy-based differential evolution operators are adopted to create aggressive offspring. The superior exploration and exploitation ability of the proposed algorithm is validated via some state-of-the-art CMOEAs over artificial benchmarks and real-world problems. The experimental results show that our proposed algorithm gained a better, or more competitive, performance than the other competitors, and it is an effective approach to balancing objective optimization and constraint satisfaction.

This paper proposes a modified version of the weighted mean of vectors algorithm (mINFO), which combines the strengths of the INFO algorithm with the Enhanced Solution Quality Operator (ESQ). The ESQ boosts the quality of the solutions by avoiding optimal local values, verifying that each solution moves towards a better position, and increasing the convergence speed. Furthermore, we employ the mINFO algorithm to optimize the connection weights and biases of feedforward neural networks (FNNs) to improve their accuracy. The efficacy of FNNs for classification tasks is mainly dependent on hyperparameter tuning, such as the number of layers and nodes. The mINFO was evaluated using the IEEE Congress on Evolutionary Computation held in 2020 (CEC’2020) for optimization tests, and ten chemical data sets were applied to validate the performance of the FNNs classifier. The proposed algorithm’s results have been evaluated with those of other well-known optimization methods, including Runge Kutta optimizer’s (RUN), particle swarm optimization (PSO), grey wolf optimization (GWO), Harris hawks optimization (HHO), whale optimization algorithm (WOA), slime mould algorithm (SMA) and the standard weighted mean of vectors (INFO). In addition, some improved metaheuristic algorithms. The experimental results indicate that the proposed mINFO algorithm can improve the convergence speed and generate effective search results without increasing computational costs. In addition, it has improved the FNN’s classification efficiency.

To address the inflexibility of the convolutional autoencoder (CAE) in adjusting the network structure and the difficulty of accurately delineating complex class boundaries in power load data, a particle swarm optimization deep clustering method (DC-PSO) is proposed. First, a particle swarm optimization algorithm for automatically searching the optimal network architecture and hyperparameters of CAE (AHPSO) is proposed to obtain better reconstruction performance. Then, an end-to-end deep clustering model based on a reliable sample selection strategy is designed for the deep clustering algorithm to accurately delineate the category boundaries and further improve the clustering effect. The experimental results show that the DC-PSO algorithm exhibits high clustering accuracy and higher performance for the power load profile clustering.

Industrial internet of things (IIoT) connects traditional industrial devices with the network to provide intelligent services, which is regarded as the key technology for achieving Industry 4.0 and enabling the transformation of the manufacturing sector. Multi-access edge computing (MEC) has brought significant opportunities to expedite the development of IIoT. However, the unique task characteristics and dense deployment of IIoT devices, coupled with the resource starvation problem (RSP) arising from the limited resources of edge servers, pose challenges to the direct applicability of existing MEC algorithms in MEC-assisted IIoT scenarios. To this end, a multi-objective evolutionary algorithm is proposed to simultaneously optimize delay and energy consumption for multi-workflow execution in resource-limited IIoT. First, the initialization of execution location based on delay and the initialization of execution order satisfying the priority constraint can generate high-quality initial solutions. Then, the improved crossover and mutation operations guide the population evolution, which can span the large infeasible solution region. Finally, dynamic task scheduling (DTS) dynamically changes the execution location of tasks affected by RSP according to the execution efficiency, so as to avoid the tasks blindly waiting for server resources. The comprehensive simulation results demonstrate the effectiveness of the proposed method in achieving a balance between the execution delay and energy consumption of IIoT devices.

With the advent of "Industry 4.0", more matrix manufacturing workshops have adopted automated guided vehicle (AGV) for material handling. AGV transportation has become a key link in manufacturing production. Traditional AGVs scheduling problem (AGVSP) is studied in depth. However, most research overlooks an important problem, in production with limited resources, the number of AGVs is insufficient. Therefore, the wait time of workstations is longer than expected. The service time of the task is delayed and the cost is increased. To solve above problem, this paper proposes the binary discrete enhanced Jaya (BDE-Jaya) algorithm. The main goal is to minimize transportation cost, including AGV traveling cost, service early penalty, and total tardiness (TTD). A key-task shift method is proposed to reduce TTD and task service early penalty. Two heuristics based on the problem features are designed to generate the initial solution. In the evolutionary stage, three offspring generation methods are used to improve the algorithm exploitation capability and exploration capability. Then, an insertion-based repair method is designed to prevent the exploitation process falling into local optimum. Furthermore, three parameters are proposed to improve the performance of the algorithm. Finally, simulation experiment shows that the proposed BDE-Jaya algorithm has significant advantages compared with other algorithms.

Constrained multimodal multi-objective optimization (CMMOPs) involves multiple equivalent constrained Pareto optimal sets (CPSs) matching the same constrained Pareto front (CPF). An essential challenge in solving CMMOPs is how to balance exploration and exploitation in searching for the CPSs. To tackle this issue, a dynamic constrained co-evolutionary multimodal multi-objective algorithm termed DCMMEA is developed in this paper. DCMMEA involves a constraint-relaxed population for handling constraints and a convergence-relaxed population for improving convergence quality. Subsequently, a constraint-relaxed epsilon strategy that considers the constraint violation degree between individuals is designed and applied dynamically in the constraint-relaxed population to develop equivalent CPSs. Similarly, a dynamic convergence-relaxed epsilon strategy that considers the differences between objective values is developed and used dynamically in the convergence-relaxed population. It explores CPSs with high convergence quality and transfers the convergence knowledge to the constraint-relaxed population. Additionally, the constraint- relaxed population size is dynamically increased and the convergence-relaxed population size is dynamically decreased to balance the exploration and exploitation procedures. Experiments are performed on standard CMMOP test suites and validate that DCMMEA obtains superior performance on solving CMMOPs in comparison to state-of-the-art algorithms. Also, DCMMEA is implemented on standard CMOPs and demonstrated good performance in handling CMOPs.