Huixiang Zhen , Bing Xue , Wenyin Gong , Mengjie Zhang , Ling Wang
{"title":"Offline evolutionary optimization with problem-driven model pool design and weighted model selection indicator","authors":"Huixiang Zhen , Bing Xue , Wenyin Gong , Mengjie Zhang , Ling Wang","doi":"10.1016/j.swevo.2025.102034","DOIUrl":"10.1016/j.swevo.2025.102034","url":null,"abstract":"<div><div>Offline data-driven evolutionary algorithms aim to provide a promising solution based on the collected historical data, without online real fitness evaluations. However, the suitability of surrogate models varies significantly across different problem types, and current research often overlooks the relationship between problem characteristics and model performance. To address this gap, we propose a novel offline data-driven evolutionary algorithm, termed MSEA, which integrates a problem-driven model pool design and a weighted indicator-based model selection mechanism. The model pool is carefully designed, incorporating four distinct surrogate models tailored for various optimization landscapes to align with diverse problem characteristics. A weighted selection indicator, derived from both model evaluation and solution quality assessment, is employed to dynamically select the most suitable model for the optimization problem. Extensive experimental results demonstrate that MSEA effectively identifies and utilizes the optimal model from the pool for specific offline optimization tasks. Compared to five state-of-the-art offline data-driven methods, MSEA achieved optimal results for 26 out of 32 functions across dimensions ranging from 10 to 100 and also exhibited faster running times. Furthermore, in high-dimensional spaces, MSEA achieved the best optimization results in dimensions ranging from 200 to 500. Our code is available at <span><span>https://github.com/zhenhuixiang/MSEA</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102034"},"PeriodicalIF":8.2,"publicationDate":"2025-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144364917","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Evolutionary algorithm with cross-diversity integration and mutation synergy operation for multi-objective recommendation","authors":"Liang Chu , Ye Tian","doi":"10.1016/j.swevo.2025.102031","DOIUrl":"10.1016/j.swevo.2025.102031","url":null,"abstract":"<div><div>Recommendation algorithms have become increasingly prevalent in modern society, addressing overload by delivering content aligned with user preferences. While accuracy is prioritized in traditional approaches, diversity is also crucial in recommendation systems. However, the balance between these two objectives is challenged by a fundamental trade-off. To address this issue, an enhanced multi-objective evolutionary algorithm (MOEA-EMRS) is proposed, in which cross-diversity mechanism and mutation synergy operation are integrated for multi-objective recommendations. MOEA-EMRS integrates three core components: a novel population initialization mechanism that constructs a distinctive primitive population with enhanced diversity and accuracy, a diversity-preserving crossover operator, and objective-oriented mutation operation specifically designed to reinforce Pareto optimality. To evaluate the algorithm’s performance, comparative experiments were conducted between MOEA-EMRS and existing multi-objective models. Experimental results demonstrate that MOEA-EMRS outperforms existing algorithms in performance effectiveness.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102031"},"PeriodicalIF":8.2,"publicationDate":"2025-06-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144335762","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Shuwei Zhu , Yimo Zhang , Wei Fang , Meiji Cui , Kalyanmoy Deb
{"title":"Regression and relation-assisted evolutionary algorithm for high-dimensional expensive multi-objective optimization","authors":"Shuwei Zhu , Yimo Zhang , Wei Fang , Meiji Cui , Kalyanmoy Deb","doi":"10.1016/j.swevo.2025.101978","DOIUrl":"10.1016/j.swevo.2025.101978","url":null,"abstract":"<div><div>Surrogate-assisted evolutionary algorithms (SAEAs) have gained a lot of attention to handle expensive multi-objective optimization problems (EMOPs). However, when it comes to high-dimensional EMOPs (HEMOPs), the performance of existing SAEAs degrades dramatically because of the dimensionality sensitivity issue, in which effective surrogate models are difficult to build. To this end, we propose a regression- and relation-assisted evolutionary algorithm (<span><math><mrow><msup><mrow><mtext>R</mtext></mrow><mrow><mn>2</mn></mrow></msup><mtext>AEA</mtext></mrow></math></span>) to deal with HEMOPs, which involves a regression-assisted weight optimization (RWO) stage and a relation-assisted multi-objective optimization (RMO) stage. To be specific, the RWO is facilitated by the problem transformation strategy and regression models. It reformulates the high-dimensional problem into a relative low-dimensional one and intends to converge to the Pareto-optimal front (PF) efficiently. Thereafter, the RMO concentrates on maintaining the population diversity with a new infill sampling criterion, which considers the optimization performance as well as the uncertainty estimated by the predicted entropy. To validate its effectiveness, we compare <span><math><mrow><msup><mrow><mtext>R</mtext></mrow><mrow><mn>2</mn></mrow></msup><mtext>AEA</mtext></mrow></math></span> with five state-of-the-art algorithms on various benchmark test suites with dimensions varying from 50 to 200, and six real-world HEMOPs. Experimental results show the superiority of <span><math><mrow><msup><mrow><mtext>R</mtext></mrow><mrow><mn>2</mn></mrow></msup><mtext>AEA</mtext></mrow></math></span> in terms of convergence speed and diversity maintenance with limited computational resources.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 101978"},"PeriodicalIF":8.2,"publicationDate":"2025-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144331060","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Wen Zhou , Xinchun Yi , Changyi Li , Zhiwei Ye , Qiyi He , Xiuwen Gong , Qiao Lin
{"title":"An improved genetic algorithm and clone selection optimization-based gated recurrent unit networks for earthquake magnitude prediction","authors":"Wen Zhou , Xinchun Yi , Changyi Li , Zhiwei Ye , Qiyi He , Xiuwen Gong , Qiao Lin","doi":"10.1016/j.swevo.2025.102023","DOIUrl":"10.1016/j.swevo.2025.102023","url":null,"abstract":"<div><div>Earthquake magnitude prediction is a vital rendezvous for human safety, economic and property losses. The earthquake occurrence process represents a highly complex nonlinear problem. Meanwhile, artificial intelligence methods have emerged as automated and intelligent frameworks for addressing magnitude prediction challenges. However, these approaches ignore redundant features and have lower prediction accuracy. Genetic Algorithms (GA) excel in feature selection and Gated Recurrent Units (GRU) have strong time series prediction capabilities. Therefore, we propose a novel earthquake magnitude prediction method, named Improved GA and a Clone Selection Optimization-based GRU (IGA-CSOGRU). First, an improved GA with generation gap strategy is presented to enhance the feature selection capability of time-series data in prediction models. Second, GRU is implemented as the core prediction model. To optimize its hyperparameters, a novel approach combining Latin hypercube sampling with adaptive mutation CSO is introduced, thereby enhancing prediction performance. Finally, to validate the performance of the proposed IGA-CSOGRU, a novel earthquake magnitude prediction dataset is constructed, which is acquired from the self-developed Acoustic & Electromagnetics to AI (AETA) platform. Evaluation metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> were used for assessment. The proposed IGA-CSOGRU model demonstrates significant performance improvements across all datasets, achieving an average RMSE reduction of 5%–7% compared to all baseline methods, highlighting the model’s superior capability in handling challenging time series prediction tasks. The implementation code supporting the findings of this study is available at <span><span>https://github.com/123fggv/Earthquake-prediction</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102023"},"PeriodicalIF":8.2,"publicationDate":"2025-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144331059","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Zhanglu Hou , Jialu Ye , Yizhang Xia , Yibin Gong , Juan Zou , Shengxiang Yang
{"title":"A decomposition framework with dual populations and dual stages for constrained multi-objective optimization","authors":"Zhanglu Hou , Jialu Ye , Yizhang Xia , Yibin Gong , Juan Zou , Shengxiang Yang","doi":"10.1016/j.swevo.2025.102030","DOIUrl":"10.1016/j.swevo.2025.102030","url":null,"abstract":"<div><div>The main challenge of constrained multi-objective optimization involves achieving a good balance among feasibility, convergence, and diversity simultaneously. However, most existing methods exhibit an imbalance, making it hard to converge towards Pareto-optimal front (PF) while maintaining the diversity of feasible solutions. To address this issue, this paper proposes a new dual-population and dual-stage constrained multi-objective optimization algorithm, denoted as DD-M2M, based on the multi-objective-to-multi-objective decomposition framework (M2M). In this approach, dual populations, referred to as <span><math><mrow><mi>C</mi><mspace></mspace><mi>P</mi></mrow></math></span> and <span><math><mrow><mi>D</mi><mspace></mspace><mi>P</mi></mrow></math></span>, are employed collaboratively. More specifically, in the first stage, the <span><math><mrow><mi>D</mi><mspace></mspace><mi>P</mi></mrow></math></span> evolves independently without considering constraints, focusing solely on convergence towards the unconstrained PF, while the <span><math><mrow><mi>C</mi><mspace></mspace><mi>P</mi></mrow></math></span> evolves with a weak collaboration with the <span><math><mrow><mi>D</mi><mspace></mspace><mi>P</mi></mrow></math></span>, driven by feasibility-based environment selection rules. In the second stage, the dual populations are both divided into multiple sub-populations within the M2M framework, generating offspring in a strongly collaborative manner to eventually converge to the constrained PF while ensuring solution diversity. Experimental results on two widely-used test suites fully demonstrate the superiority of DD-M2M compared to seven state-of-the-art methods on most test problems. Additionally, the proposed method is applied to real-world problems, and experimental results confirm its effectiveness in addressing practical challenges.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102030"},"PeriodicalIF":8.2,"publicationDate":"2025-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144322625","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Balancing objective and switch cost for robust optimization over time","authors":"Zhening Liu, Handing Wang","doi":"10.1016/j.swevo.2025.102022","DOIUrl":"10.1016/j.swevo.2025.102022","url":null,"abstract":"<div><div>Dynamic optimization problems (DOPs) commonly arise in real-world scenarios where objective functions and constraints change over time. While existing methods primarily focus on tracking optimal solutions across different environments, the cost of switching solutions cannot be overlooked in practical applications. Robust optimization over time (ROOT) addresses this challenge by simultaneously maximizing objective values and minimizing solution switch costs. This paper proposes a novel ROOT method, introducing a dynamic balancing mechanism that adjusts the search direction according to the correlation between objective value and switch cost. Additionally, a robust solution selection strategy is developed, utilizing switch cost as a constraint to pre-screen solutions before selecting the optimal robust solution. Comprehensive experiments on ROOT test problems of varying dimensions and switch cost weights validate the effectiveness of the proposed approach. Comparisons with ROOT algorithms, dynamic evolutionary algorithms, multi-objective evolutionary algorithms and single-objective evolutionary algorithms demonstrate that the proposed method achieves superior overall performance. Furthermore, ablation studies confirm the effectiveness of the balancing mechanism and the robust solution selection strategy in enhancing optimization quality.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102022"},"PeriodicalIF":8.2,"publicationDate":"2025-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144331058","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A knowledge transfer-based co-evolutionary algorithm for dynamic large-scale crude oil scheduling","authors":"Wanting Zhang, Wenli Du, Wei Du","doi":"10.1016/j.swevo.2025.102024","DOIUrl":"10.1016/j.swevo.2025.102024","url":null,"abstract":"<div><div>Confronted with the intricacies arising from expanding production scales and numerous uncertain events, large-scale dynamic scheduling has emerged as a practical approach in industries. However, within the framework of predictive–reactive scheduling, few works have explored the trade-off between the optimality and stability of the solutions generated. To fill this gap, this paper investigates dynamic scheduling applied to the large-scale crude oil scheduling problem, which is critical for the petroleum industry. Specifically, we develop a model that incorporates multiple sources of uncertainty: vessel arrival delays; tank malfunctions; fluctuations in feed flowrates to the distillation columns; and intermediate product demand. To solve this problem effectively, a knowledge transfer-based cooperative co-evolutionary algorithm (KT-CCEA) is proposed, where specific knowledge from the predictive stage is transferred to the reactive stage. Specifically, multiple subpopulations are generated around distinct reactive points, evolving across diverse search dimensions, to balance optimality and stability. Discretized differential operators are designed to overcome the limitations of standard evolutionary operators in integer-coded matrix representation. Empirical results over a set of 15 benchmark instances validate the superiority of the proposed KT-CCEA over four state-of-the-art algorithms (RSCO-SAGA, VLCEA, DMDE, and RCI-PSO). Ablation experiments on seven algorithm variants further confirm the efficacy of its core components.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102024"},"PeriodicalIF":8.2,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144313795","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"LABCAT: Locally adaptive Bayesian optimization using principal-component-aligned trust regions","authors":"E. Visser, C.E. van Daalen, J.C. Schoeman","doi":"10.1016/j.swevo.2025.101986","DOIUrl":"10.1016/j.swevo.2025.101986","url":null,"abstract":"<div><div>Bayesian optimization (BO) is a popular method for optimizing expensive black-box functions. BO has several well-documented shortcomings, including computational slowdown with longer optimization runs, poor suitability for non-stationary or ill-conditioned objective functions, and poor convergence characteristics. Several algorithms have been proposed that incorporate local strategies, such as trust regions, into BO to mitigate these limitations; however, none address all of them satisfactorily. To address these shortcomings, we propose the LABCAT algorithm, which extends trust-region-based BO by adding a rotation aligning the trust region with the weighted principal components and an adaptive rescaling strategy based on the length-scales of a local Gaussian process surrogate model with automatic relevance determination. Through extensive numerical experiments using a set of synthetic test functions and the well-known COCO benchmarking software, we show that the LABCAT algorithm outperforms several state-of-the-art BO and other black-box optimization algorithms.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 101986"},"PeriodicalIF":8.2,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144297892","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jiansheng Liu , Haoran Hu , Zhiyong Liu , Zan Yang , Liming Chen , Xiwen Cai
{"title":"Expensive constrained multi-objective optimization via adaptive surrogate-assisted dense weight multi-objective evolutionary algorithm","authors":"Jiansheng Liu , Haoran Hu , Zhiyong Liu , Zan Yang , Liming Chen , Xiwen Cai","doi":"10.1016/j.swevo.2025.102033","DOIUrl":"10.1016/j.swevo.2025.102033","url":null,"abstract":"<div><div>Expensive constrained multi-objective optimization problems (ECMOPs) face challenges in obtaining excellent results for complex <em>PF</em> shapes within limited costly evaluations efficiently and balancing the optimizing on constraints and objectives. Also, one surrogate typically cannot provide the consistent predictive abilities for multiple objectives or constraints with diverse features. This paper designs an adaptive surrogate-assisted dense weight multi-objective evolutionary algorithm (ASDWMOEA), where efficient dense weight-based dual-population evolution and effective surrogate switch mechanism are integrated. Specifically, when there is no feasible solution in the population, the algorithm ignores the constraints of the problem and uses Kriging surrogate model to optimize only for the objective of the problem. When the population enters the feasible domain, the algorithm uses the association information of external weights to implement three mutation operations for generating a large set of high-quality candidate solutions. Each external weight is then refined to produce multiple internal weights, and an elite subset of the candidate solutions is selected to form the internal population corresponding to these internal weights. Subsequently, sequential global and local searches are conducted on the internal population, and the elite individual with the most significant improvement is selected for each internal weight. Hence, the external population is updated based the two metric-based selection strategy, and the algorithm adaptively switches between the employed Kriging and RBF surrogate models based on updates to the external population. Finally, the algorithm is evaluated against seventeen advanced and latest algorithms using five test suites. The results demonstrate that ASDWMOEA exhibits strong performance.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102033"},"PeriodicalIF":8.2,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144291274","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Shuxiang Li , Yongsheng Pang , Zhaorong Huang , Xianghua Chu
{"title":"An offline-online learning framework combining meta-learning and reinforcement learning for evolutionary multi-objective optimization","authors":"Shuxiang Li , Yongsheng Pang , Zhaorong Huang , Xianghua Chu","doi":"10.1016/j.swevo.2025.102037","DOIUrl":"10.1016/j.swevo.2025.102037","url":null,"abstract":"<div><div>Many multi-objective evolutionary algorithms (MOEAs) have been proposed in addressing the multi-objective optimization problems (MOPs). However, the performance of MOEAs varies significantly across various MOPs and there is no single MOEA that performs well on all MOP instances. In addition, existing methods for adaptive MOEA selection still face limitations, which restrict the further optimization for MOPs. To fill these gaps and improve the efficiency of solving MOPs, this study proposes an offline-online learning framework combining meta-learning and reinforcement learning (O<sup>2</sup>-MRL). Instead of proposing a new MOEA or optimizing a strategy, O<sup>2</sup>-MRL solves MOPs by taking full advantage of the existing MOEAs and addresses the limitations of existing MOEA selection methods. O<sup>2</sup>-MRL can adaptively select the appropriate MOEAs for various types of MOPs with different dimensions (Offline) and automatically schedule the selected MOEAs during the optimization process (Online), offering a new idea for optimizing MOPs. To evaluate the performance of the proposed O<sup>2</sup>-MRL, forty-seven benchmark MOPs are used as instances, and nine representative MOEAs are selected for comparison. Comprehensive experiments demonstrate the significant efficiency of O<sup>2</sup>-MRL, as it achieves optimal solutions in 60.28 % of the MOPs across different dimensions and improves the optimization results in 48.23 % of them, with an average improvement of 8.72 %. In addition to maintaining high optimization performance, O<sup>2</sup>-MRL also demonstrates superior convergence speed and stability across various types of MOPs. Two real-world MOPs are employed to evaluate the practicality of O<sup>2</sup>-MRL, and the experimental results indicate that it achieves optimal solutions in both cases.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"97 ","pages":"Article 102037"},"PeriodicalIF":8.2,"publicationDate":"2025-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144279770","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}