A paper published in "Journal of Hydrology" 2021 highly recommended using our HARD-DE algorithm as the calibration method for the air2water model in future studies. Many thanks to Prof. Adam Piotrowski for bringing this to my attention.
The PaDE paper has been a highly cited paper since Nov 2019.
Parts of the publications are listed below:
Meng, Zhenyu* and Junyuan Zhang. "QUATRE-EMS: QUATRE algorithm with novel adaptation of evolution matrix and selection operation for numerical optimization." Information Sciences, 651(2023): 119714 (Paper/Code). (Paper in Evolutionary Computation)
Comments on QUATRE-EMS: A state-of-the-art QUATRE variant, namely QUATRE-EMS algorithm, was proposed for numerical optimization. There are three innovations involved in the new DE variant: First, the first innovation is an extension of the QUATRE algorithm, the inborn weakness of DE caused by employing crossover rate CR is tackled by employing the novel evolution matrix M; Second, a novel adaptation of evolution matrix is proposed in the QUATRE-EMS algorithm aiming at better adaptation of the landscape of the objective during evolution; Third, a new selection operation was employed instead of employing the canonical selection operation in DE algorithm. A large test suite containing 100 benchmarks is employed in algorithm validation. The results support the superiority of our QUATRE-EMS algorithm.
QUATRE-EMS论文介绍：本文提出了一种叫QUATRE-EMS的新QUATRE方法。如同这个新名字一般，希望QUATRE算法在求解数值优化问题时能够如同EMS快递般精准到达。QUATRE-EMS算法相比经典的QUATRE算法进行了如下改进：1. 延续了QUATRE的方式，采用进化矩阵克服差分进化中与生俱来的解空间搜索偏见，该偏见是因为使用交差率CR导致的；2. 进一步增强了进化矩阵更新方式，以期实现对目标函数不同landscape能够更加有效适应；3. 采用了一种新的选择方式，增强了种群多样性。为验证QUATRE-EMS的整体优化效果，我们采用了包含100个国际常用函数的大测试集合。实验表明QUATRE-EMS方法效果优异。
Meng, Zhenyu* and Yuxin Chen. "Differential Evolution with exponential crossover can also be competitive on numerical optimization." Applied Soft Computing, 146(2023):110750 (Paper/Code). (Paper in Evolutionary Computation)
Comments on DE-EXP: The first powerful DE variant with exponential crossover, namely the DE-EXP algorithm, which is superior to the recent winner DE variants, e.g. LSHADE, iLSHADE and jSO, in Congress on Evolutionary Computation (CEC) competitions, is developed for numerical optimization. The majority of DE researchers believe that DE variants employing binomial crossover usually obtain superior performance than the ones employing exponential crossover on numerical optimization and DE variants with exponential crossover are good at tackling optimization problems with linkages among neighboring variables. On the contrary, here in this paper, a new perspective is proposed that (1) the commonly used exponential crossover and binomial crossover actually implement the same effect, selecting a certain vertex from the hyper-cube; (2) from this perspective of view, both the binomial crossover and exponential crossover have inborn weakness and can be transformed into each other after finding the proper crossover rate Cr for a certain number of generations, and the transformation can be implemented by calculating the evolution matrix in our QUATRE algorithm; (3) The proper Cr values in exponential crossover were usually in a much narrower range than in binomial crossover when obtaining equivalent performance, and finding these proper Cr values was very difficult for a DE variant employing exponential crossover let alone finding its corresponding parameter control; (4) There always exists a framework on which the exponential crossover and the binomial crossover can be transformed into each other with equivalent performance, and beat all the recent state-of-the-art DE algorithms. To discover such a framework is one of the future work in our team. Besides these, there are still some contributions in our DE-EXP algorithm summarized as follows: (1) A novel parameter control of crossover rate is developed for exponential crossover and the value of Cr can be automatically generated not only in the initialization stage but also during the evolution. (2) A black-box model illustrating the fitness-value-dependency weakness of the recent winner DE variants in CEC competitions is given and a novel fitness-value-independent adaptation scheme, the dimension improvement based adaptation, for scale factor is proposed in the DE-EXP algorithm to overcome the fitness-value-dependency weakness.
DE-EXP论文介绍：本文提出了一种指数交叉的差分进化算法来解决数值优化问题，该算法是DE领域近十多年来的第一个能够打败LSHADE, jSO等先进算法的基于指数交叉的差分进化算法。与差分进化领域绝大部分研究者所认为的“二项交叉相较于指数交叉更适用于数值优化问题，指数交叉更适用于维度之间有关联的优化问题”不同，我们认为“1. 从空间搜索角度，指数交叉与二项交叉都是实现高维立方体的选顶点操作，这两种交叉方式都存在解空间搜索偏见，该偏见在我们提的QUATRE算法中有详细论述。2.无论维度之间是否存在联系，只要找到合适的参数控制方案，指数交叉可以获得与二项交叉同等的优化效果。只是相比二项交叉，指数交叉的控制参数可选范围远小于二项交叉，故而找到合适的参数控制方案十分困难；3. 指数交叉与二项交叉可以通过求QUATRE算法中的进化矩阵的方式实现相互转换；4.肯定存在一个框架，在该框架下无论使用指数交叉亦或使用二项交叉能取得等价的优化效果，且可以打败当前最先进的差分进化算法”。在DE-EXP算法中，还引入了以下创新点：1. 指数交叉的交叉率Cr不用赋初始值，自始至终自动生成；2. 引入了基于维度改进的参数控制方案，解决了目标函数适应值的依赖问题。
Meng, Zhenyu*. "Dimension improvements based adaptation of control parameters in Differential Evolution: a fitness-value-independent approach." Expert Systems with Applications 223 (2023):119848 (Paper/Code). (Paper in Evolutionary Computation)
Comments on PaDE-pet: Nearly all the recently proposed state-of-the-art DE variants for numerical optimization employed the fitness improvement based weight in the adaptation of control parameters, obviously, these algorithms can only tackle the optimization scenario that the exact fitness values are available. However, there are many optimization cases in the real-world applications that the exact fitness value is unavailable (or deliberately hidden). Therefore, a novel idea of the fitness-independent DE algorithms is proposed in the paper. Dimension improvement based weight in the adaptation of control parameters is proposed rather than employing the former fitness improvement based weight in the adaptation of control parameters. The motivation for the fitness improvement based weight is transparent, the bigger the fitness improvement is, the larger the weight is. Accordingly, the motivation of the dimension improvement based weight is also transparent and much better from the population diversity perspective. The bigger the standard deviation of the dimensional improvement in the offspring is, the larger the weight is. This adaptation can help the diversity increment of the population and to some extent avoid premature convergence, consequently, be more likely to obtain excellent performance.
PaDE-pet论文介绍：本文首次提出了不依赖目标函数适应值的差分进化算法这一概念，并给出了区分依赖和不依赖目标函数适应值的问题模型。正如TDE的介绍中所描述的“准确的目标函数适应值并不是差分进化算法执行的必要条件，能够区分输入向量间的优劣才是差分进化执行的必要条件”。最原始的差分进化算法并不存在对目标函数适应值的依赖问题，近年来的差分进化算法对目标函数适应值的依赖源于基于目标函数适应值改进的参数控制方案中。基于目标函数适应值改进的参数控制方案始于姚新老师团队提出的rJADE算法，经由LSHADE发扬光大。虽然这类参数控制方案优势十分明显，但缺陷也十分突出，即不适用于无精确的目标函数适应值的优化场景。在现有文献中，不采用基于目标函数适应值改进的参数控制方案的差分进化算法均不存在依赖目标函数适应值的问题，但绝大多数优化效果均不理想。能够有效代替基于目标函数适应值改进的参数控制方案仅有Adam等人提出的基于距离的参数控制方案，但Adam等人并未明确“不依赖目标函数适应值”这一分类。在PaDE-pet中，我们提出了一种基于维度改进的参数控制，该参数控制不同于基于距离的参数控制方案，“维度改进的方差”这一统计量被用作控制参数的权重。由于基于维度改进的参数控制方案只需要计算有改进的维度，故而计算复杂度低于基于距离的参数控制方案，同时在通用测试集下的对比结果也凸显了该基于维度改进的差分进化算法的优势。由于基于维度改进的参数控制方案给后代中那些维度变动大的优秀个体赋予更高权重值，这种参数控制方式在一定程度上能够增加种群中的个体位置多样性，从而避免过速收敛，这也更易于提高整体优化效果。
Song, Zhenghao, Chongle Ren, Zhenyu Meng*. "Parameter Identification of Photovoltaic Models Using an Improved Differential Evolution With Selective Perturbation." IEEE Transactions on Industrial Informatics 21, no 4 (2025): 2908 - 2916 (Paper/Code). (Paper in Evolutionary Computation)
Wang, Xiang, Hao Dou, Dibo Dong, Zhenyu Meng*. "Graph anomaly detection based on hybrid node representation learning." Neural Networks 185(2025): 107169 (Paper/Code). (Paper in Graph Neural Networks)
Yu, Laiqi, Zhenyu Meng*. "Surrogate-assisted Differential Evolution: A Survey." Swarm and Evolutionary Computation 94 (2025): 101879 (Paper/Code). (Paper in Evolutionary Computation)
Xu, Qiutong, Zhenyu Meng*. "Differential Evolution with multi-stage parameter adaptation and diversity enhancement mechanism for numerical optimization." Swarm and Evolutionary Computation 92(2025): 101829 (Paper/Code). (Paper in Evolutionary Computation)
Zhou, Wencan, Zhenyu Meng*. "An adaptive differential evolution with dynamic perturbation and dimensional bidirectional crossover mechanism for diversity enhancement." Engineering Applications of Artificial Intelligence 141(2025): 109750 (Paper/Code). (Paper in Evolutionary Computation)
Yu, Laiqi, Zhenyu Meng*, Haibin Zhu. "A Hierarchical Surrogate-Assisted Differential Evolution with Core Space Localization." IEEE Transactions on Cybernetics 55, no 2 (2025): 939-952 (Paper/Code). (Paper in Evolutionary Computation)
Ren, Chongle, Zhenyu Meng*. "A survey on expensive optimization problems using differential evolution." Applied Soft Computing 170 (2025): 112727 (Paper/Code). (Paper in Evolutionary Computation)
Yu, Laiqi, Zhenyu Meng*. "Surrogate-Assisted Differential Evolution with multiple sampling mechanisms for high-dimensional expensive problems." Information Sciences 687 (2025): 121408 (Paper/Code). (Paper in Evolutionary Computation)
Li, Juncan, Zhenyu Meng*, "Elite-guided resampling and multi-mutation based differential evolution with exponential crossover for numerical optimization." Expert Systems with Applications 258 (2024): 125159 (Paper/Code). (Paper in Evolutionary Computation)
Ren, Chongle, Qiutong Xu, Zhenyu Meng*, Jeng-Shyang Pan. "Surrogate-assisted fully-informed particle swarm optimization for high-dimensional expensive optimization." Applied Soft Computing 167(2024):112464 (Paper/Code). (Paper in Evolutionary Computation)
Zhang, Quanbin, Zhenyu Meng*. "HAPI-DE: Differential evolution with hierarchical archive based mutation strategy and promising information." Swarm and Evolutionary Computation 91 (2024): 101705 (Paper/Code). (Paper in Evolutionary Computation)
Ren, Chongle, Zhenghao Song, Zhenyu Meng*. "Photovoltaic model parameters identification using diversity improvement-oriented differential evolution." Swarm and Evolutionary Computation 90 (2024): 101689 (Paper/Code). (Paper in Evolutionary Computation)
Lin, Xin, Zhenyu Meng*. "Surrogate-assisted evolutionary framework with an ensemble of teaching-learning and differential evolution for expensive optimization." Information Sciences 680 (2024): 121137(Paper/Code). (Paper in Evolutionary Computation)
Wang, Xiang, Weikang Deng, Zhenyu Meng*, Dewang Chen. "Hybrid-attention mechanism based heterogeneous graph representation learning." Expert Systems with Applications 250 (2024): 123963 (Paper/Code). (Paper in Graph Neural Networks)
Meng, Zhenyu*, Xin Lin, Dewang Chen. "ACD-DE: An adaptive cluster division Differential Evolution for mitigating population diversity deficiency." Information Sciences 679 (2024): 121091(Paper/Code). (Paper in Evolutionary Computation)
Lin, Xin, Zhenyu Meng*. "An adaptative differential evolution with enhanced diversity and restart mechanism." Expert Systems with Applications 249 (2024): 123634 (Paper/Code). (Paper in Evolutionary Computation)
Ren, Chongle, Zhenghao Song, Zhenyu Meng*. "Differential Evolution with fitness-difference based parameter control and hypervolume diversity indicator for numerical optimization." Engineering Applications of Artificial Intelligence 133 (2024): 108081 (Paper/Code). (Paper in Evolutionary Computation)
Li, Juancan, Zhenyu Meng*. "Global Opposition Learning and Diversity ENhancement based Differential Evolution with exponential crossover for numerical optimization." Swarm and Evolutionary Computation 87 (2024): 101577 (Paper/Code). (Paper in Evolutionary Computation)
Song, Zhenghao, Chongle Ren, Zhenyu Meng*. "An adaptive differential evolution with opposition-learning based diversity enhancement." Expert Systems with Applications 243 (2024): 122942 (Paper/Code). (Paper in Evolutionary Computation)
Yu, Laiqi, Chongle Ren, Zhenyu Meng*. "A Surrogate-Assisted Differential Evolution with fitness-independent parameter adaptation for high-dimensional expensive optimization." Information Sciences 662 (2024): 120246 (Paper/Code). (Paper in Evolutionary Computation)
Song, Zhenghao, Chongle Ren, Zhenyu Meng*. "Differential Evolution with perturbation mechanism and covariance matrix based stagnation indicator for numerical optimization." Swarm and Evolutionary Computation 84 (2024): 101447 (Paper/Code). (Paper in Evolutionary Computation)
Meng, Zhenyu*, Quanbin Zhang. "HPDE: A dynamic Hierarchical Population based Differential Evolution with novel diversity metric." Engineering Applications of Artificial Intelligence 126 (2023): 106989 (Paper/Code). (Paper in Evolutionary Computation)
Meng, Zhenyu*, Zhenghao Song, Xueying Shao, Junyuan Zhang, Huarong Xu, "FD-DE: Differential Evolution with fitness deviation based adaptation in parameter control." ISA Transactions 139 (2023): 227-290 (Paper/Code). (Paper in Evolutionary Computation)
Zhang, Quanbin, Zhenyu Meng*. "Adaptive differential evolution algorithm based on deeply-informed mutation strategy and restart mechanism." Engineering Applications of Artificial Intelligence 126 (2023): 107001 (Paper/Code). (Paper in Evolutionary Computation)
Song, Zhenghao, Zhenyu Meng*. "Differential Evolution with wavelet basis function based parameter control and dimensional interchange for diversity enhancement." Applied Soft Computing 144 (2023): 110492 (Paper/Code). (Paper in Evolutionary Computation)
Meng, Zhenyu*, Cheng Yang. "Two-stage Differential Evolution with novel parameter control." Information Sciences 596 (2022): 321-342 (Paper/Code). (Paper in Evolutionary Computation)
Comments on TDE: a novel two-stage DE algorithm, the tDE algorithm, was proposed in this paper, and each stage employed a unique mutation strategy: a novel historical-solution-based mutation strategy was employed in the first stage while the commonly used mutation strategy "DE/target-to-pbest/1/bin with external archive" was used in the second. By incorporating the historical solutions instead of only employing the ones in the current generation, the novel mutation strategy in the first stage of the evolution can get better perception of the landscape of the objective, consequently, a better exploration in the earlier stage of the evolution can be secured by our algorithm. Moreover, the Novel fitness-value-independent Parameter Control, called the NPC technique, proposed in our former work, was also employed in tDE, which can help the algorithm obtain excellent performance and cater to a larger optimization scenario especially those that the fitness values are unavailable.
TDE论文介绍：本文提出了一种分段策略的差分进化算法，进化过程中的两个阶段采用单独的变异策略，即：算法在第一阶段的执行过程中采用一种基于历史解的变异策略，在第二阶段执行过程中采用近年来常用的基于外部存储的“DE/target-to-pbest/1/bin”策略。我们之前的研究发现，进化过程中时间维度上种群个体的分布特征可以在一定程度上反应空间维度上目标函数的结构特征。第一阶段的变异策略通过引入历史个体所反映的目标函数结构特征来实现更好的空间搜索（exploration），第二阶段在基于第一阶段的基础上采用近年来优化竞赛中获胜的基于外部存储的“DE/target-to-pbest/1/bin”策略完成种群进化过程。此外，本文采用了本团队之前研究中提出的“不依赖目标函数适应值的参数控制方案”（NPC技术）来改进测试向量的质量。通过以上各部分的提出，该TDE算法的优化效果不但比现有文献中绝大多数先进（state-of-the-art）差分进化算法还要好，而且该算法还能解决比这些先进算法还要广的尤其是那些目标函数适应值不可知的优化场景。这里要强调一下，准确的目标函数适应值并不是差分进化算法执行的必要条件，“能够区分输入向量之间的优劣”才是差分进化算法执行的必要条件。
Meng, Zhenyu*, Yuxin Zhong, Guojun Mao, Yan Liang, "PSO-sono: A novel PSO variant for single-objective numerical optimization." Information Sciences 586 (2022): 176-191 (Paper/Code). (Paper in Swarm Intelligence)
Comments on PSO-sono: a novel PSO variant is proposed for single-objective numerical optimization. In the algorithm, a sorted particle swarm with hybrid paradigms, novel parameter adaptation schemes and novel fully-informed search of particles are proposed to enhance the overall performance of PSO for single-objective numerical optimization. Experiment results under 88 benchmarks from CEC2013, CEC2014 and CEC2017 test suites show the superiority of our PSO-sono algorithm.
PSO-sono论文介绍：本文提出了一种新颖的用于解决单目标实参优化的PSO-sono算法，该算法采用一种基于排序种群的混合迭代策略及与其对应的参数适应调整机制和全局感知粒子搜索策略。通过上述三个部分的提出，PSO算法在求解单目标实参优化问题的整体能力得到了显著增强，三个通用测试集共88个评测函数验证了该PSO-sono算法出色的整体优化效果。
Meng, Zhenyu*, Yuxin Zhong, Cheng Yang, "CS-DE: Cooperative Strategy based Differential Evolution with Population Diversity Enhancement." Information Sciences 577 (2021):663-696 (Paper/Code). (Paper in Evolutionary Computation)
Comments on CS-DE: Two similar cooperative strategies sharing the same parameter control with different advantages are employed in the evolution, and the main difference between the two strategies lies in the external archive. The external archive in one mutation strategy records only the inferior solutions while the external archive in the other records the historical solutions. Generally, the individual distribution over the time series evolution can to some extent reflects the landscape of the objective. By incorporating historical solutions, the mutation strategy can get a better perception of the landscape of the objective. Moreover, the incorporation of inferior solutions can diversify the trial vectors which usually leads to better optimization performance. In the CS-DE algorithm, the individuals of the population can choose a mutation strategy in an adaptive manner, which can make full use of both the advantages of the two mutation strategies. Furthermore, a novel diversity indicator was proposed and diversity enhancement can be launched if necessary which can draw the population out of some local optima. Therefore, overall better performance can be secured by our CS-DE algorithm and the experiment results under the commonly used 88 benchmarks from CEC2013, CEC2014 and CEC2017 support its superiority.
CS-DE算法介绍：本文提出了一种协同进化机制的差分进化算法，该算法提出了两种极其相似、可以共享参数控制方案但各有优势的变异策略。这两种策略的主要差别在于外部存储，一种只存储进化过程中的次等解，另一种存储进化过程中的历史解。不同于以往的变异策略只用当前代信息产生变异向量，基于外部存储的“DE/target-to-pbest/1/bin”变异策略首次考虑到进化过程中信息的有效性，把进化过程中的次等解引入到变异向量生成中来，该变异策略是我们CS-DE算法中的一种策略。除此之外，在之前的研究中我们发现，进化过程中时间维度上种群个体的分布特征可以在一定程度上反映空间维度上目标函数的结构特征。把历史解引入变异策略可以更好的感知目标函数结构特征，从而实现更好地空间搜索，该策略是我们的第二种变异策略。此外，本文还提出了一种种群多样性检测指标，当进化过程中的多样性检测指标满足特定条件后将进行种群多样性增强，进而实现更好的整体优化效果。
Meng, Zhenyu*, Cheng Yang, "Hip-DE: Historical population based mutation strategy in Differential Evolution with parameter adaptive mechanism." Information Sciences 562 (2021): 44-77 (Paper/Code). (Paper in Evolutionary Computation)
Comments on Hip-DE: In this paper, a novel historical population-based DE algorithm, the Hip-DE algorithm, was proposed for numerical optimization. Generally, the majority of the mutation strategies in the literature are based on the employment of the current individuals/solutions, and the excellent mutation strategy "DE/target-to-pbest/1/bin with external archive" pointed out the advantage of incorporating (historical) inferior solutions. In this research, we further investigated the effectiveness of historical individuals during evolution. We found that the individual distribution over the time series evolution can to some extent reflects the landscape of the objective. By incorporating the historical solutions, the mutation strategy can get a better perception of the landscape of the objective, accordingly, a corresponding parameter control was also designed for the specific mutation strategy. The experiment results under a large test suite containing 88 benchmarks from CEC2013, CEC2014 and CEC2017 test suites supported the superiority of our algorithm.
Hip-DE算法介绍：本文提出了一种在变异策略中引入种群历史信息的差分进化算法，用以解决复杂的单目标实参优化问题。一般而言，现有文献中的绝大多数变异策略仅用当前种群信息来计算变异向量，显然这些变异策略都忽略了对种群历史信息的利用。基于外部存储的“DE/target-to-pbest/1/bin”变异策略首次考虑到进化过程中信息的有益性，并把进化过程中的次等解引入到变异向量生成中来，实现了非常好的优化效果。但该变异策略只用到了历史种群中的次等解，忽视了种群进化过程中的其他有益信息。我们发现，进化过程中时间维度上种群个体的分布特征可以在一定程度上反映空间维度上目标函数的结构特征。本文的Hip-DE算法通过将进化过程中的历史信息引入变异策略生成过程，更好地感知到了进化过程各阶段的目标函数结构特征，进而实现了更好的优化效果。实验数据验证了我们的Hip-DE算法的出色效果。
Meng, Zhenyu*, Yuxin Chen, Xiaoqing Li and Fang Lin, "PaDE-NPC: Parameter adaptive Differential Evolution with novel parameter control for single-objective optimization." IEEE Acccess 8 (2020): 139460 - 139478 . DOI: 10.1109/ACCESS.2020.3012885 (Paper/Code). (Paper in Evolutionary Computation)
Meng, Zhenyu*, Yuxin Chen, Xiaoqing Li, Cheng Yang and Yuxin Zhong, "Enhancing QUasi-Affine TRansformation Evolution (QUATRE) with adaptation scheme on numerical optimization." Knowledge-Based Systems 197C (2020) 105908. DOI: 10.1016/j.knosys.2020.105908 (Paper/Code) (Paper in Evolutionary Computation)
Meng, Zhenyu*, Yuxin Chen, and Xiaoqing Li. "Enhancing Differential Evolution with Novel Parameter Control." IEEE Access 8 (2020): 51145-51167 DOI: 10.1109/ACCESS.2020.2979738 (Paper/Code) (Paper in Evolutionary Computation)
Comments on DE-NPC algorithm: A novel parameter control (NPC) technique was first proposed in the algorithm, therefore, we name it the DE-NPC algorithm. In the NPC technique, a novel fitness-independent adaptation framework was proposed for the control of the distribution of control parameters F and CR. Moreover, the control parameters F and CR of each individual also employ a self-adaptive scheme including mutation and selection as jDE besides the control of their distributions. Besides these, a combined parabolic and linear population size reduction scheme was employed in the DE-NPC algorithm, which can obtain a better perception of the landscape of the objectives at the early stage of the evolution while maintaining a good balance between exploitation and exploration. The DE-NPC is verified under a large number of benchmarks from CEC2013, CEC2014 and CEC2017 test suites for real-parameter single objective optimization, and the experiment results validate its superiority not only from the optimization accuracy but also from the convergence speed perspective of view.
Meng, Zhenyu*, Cheng Yang, Xiaoqing Li, and Yuxin Chen. "Di-DE: Depth information based Differential Evolution with adaptive parameter control for numerical optimization." IEEE Access 8 (2020): 40809-40827 DOI: 10.1109/ACCESS.2020.2976845 (Paper/Code) (Paper in Evolutionary Computation)
Meng, Zhenyu*, Jeng-Shyang Pan and Kun-Kun Tseng. "PaDE: An enhanced Differential Evolution algorithm with novel control parameter adaptation schemes for numerical optimization." Knowledge-Based Systems 168C (2019): 80-99 DOI: 10.1016/j.knosys.2019.01.006 (ESI highly cited Paper/ Code) (Paper in Evolutionary Computation)
Comments on PaDE algorithm: The time stamp-based mutation strategy was deeply analyzed in the paper, this new mutation strategy can be considered as a balance between mutation strategy with an external archive and without an external archive. Furthermore, novel control parameter adaptation schemes of DE were also proposed to improve some former schemes in state-of-the-art DE variants. This work was submitted earlier than the work of HARD-DE, nevertheless, it was published later because of the longer rounds of review.
Meng, Zhenyu*, and Jeng-Shyang Pan. "HARD-DE: Hierarchical ARchive based mutation strategy with Depth information of evolution for the enhancement of Differential Evolution on numerical optimization." IEEE Access 7 (2019): 12832-12854 DOI: 10.1109/ACCESS.2019.2893292 (Paper/ Code) (Paper in Evolutionary Computation)
Comments on HARD-DE algorithm: Depth information of evolution was first incorporated into the mutation strategy, and this depth information was derived from an external hierarchical archive. New parameter adaptation schemes for the novel mutation strategy were also advanced for numerical optimization. This new HARD-DE algorithm can avoid premature convergence on some multi-modal functions in comparison with some state-of-the-art DE variants.
Meng, Zhenyu*, and Jeng-Shyang Pan. "QUasi-Affine TRansformation Evolution with External ARchive (QUATRE-EAR): an enhanced structure for differential evolution." Knowledge-Based Systems 155C (2018): 35-53. DOI: 10.1016/j.knosys.2018.04.034 (Paper/ Code) (Paper in Evolutionary Computation)
Comments on QUATRE-EAR algorithm: The former QUATRE algorithm proposed a new approach to tackling representative bias in DE algorithm with fixed Cr value, and as we know adaptive control parameters often obtain overall better optimization performance, accordingly, a novel adaptation scheme of evolution matrix M was proposed in the QUATRE-EAR algorithm aiming at obtaining better perception of objective function during the evolution. Furthermore, the only control parameter, the scale factor F, was also dynamically changed according to an adaptation scheme in the QUATRE-EAR.
Meng, Zhenyu*, Jeng-Shyang Pan, and Lingping Kong. "Parameters with adaptive learning mechanism (PALM) for the enhancement of differential evolution." Knowledge-Based Systems 141C (2018): 92-112. DOI: 10.1016/j.knosys.2017.11.015 (Paper/ Code) (Paper in Evolutionary Computation)
Comments on PALMDE algorithm: The misleading interaction was defined in the paper that a bad F and a good Cr might produce a good trial vector then consequently the bad F was treated as the right value in the adaptation of F and vice verse. In this paper, a new PALM mechanism was proposed for the independent adaptations of different control parameters. Furthermore, a time-stamp-based external archive storing inferior solutions was first proposed.
Meng, Zhenyu*, Jeng-Shyang Pan, and Huarong Xu. "QUasi-Affine TRansformation Evolutionary (QUATRE) algorithm: a cooperative swarm based algorithm for global optimization." Knowledge-Based Systems 109C (2016): 104-121. DOI: 10.1016/j.knosys.2016.06.029 (Paper/ Code) (Paper in Evolutionary Computation)
Comments on QUATRE algorithm: A cooperative evolution structure was proposed in this paper, and better performance can be obtained by increasing the number of individuals in the population with the maximum number of function evaluations unchanged, for example, the setting PS=1000 in QUATRE performed much better than PS=100 under the fixed maximum number of function evaluations, nfe_max = 100,000*D. This characteristic is different from Differential Evolution (DE). Further research found that the QUATRE structure implemented a much more reasonable search of the solution space in comparison with DE (As is known to all that exponential crossover in DE algorithm existed positional/representative bias, and the binomial crossover was asserted that it had tackled the representative bias by treating each parameter independently (from 2-dimensional view). Nevertheless, bias still existed in the binomial crossover with fixed control parameter Cr from a higher dimensional perspective of view. That's why the QUATRE paper was proposed. By introducing a new evolutionary structure, the QUATRE algorithm presented a more scientific way to select potential trial vector candidates from the hyper-cube and consequently obtained an overall better performance.). Furthermore, the DE structure can be considered as the special case of QUATRE, and the structural bias of DE can be easily found and tackled in the QUATRE structure.
Meng, Zhenyu*, and Jeng-Shyang Pan. "Monkey king evolution: a new memetic evolutionary algorithm and its application in vehicle fuel consumption optimization." Knowledge-Based Systems 97C (2016): 144-157. DOI: 10.1016/j.knosys.2016.01.009 (Paper/ Code)
Comments on MKE algorithm: Three versions were proposed in the paper, and they aimed at tackling lower dimensional (<5 D), median dimensional (5-10 D) and higher dimensional (10-100 D) black-box numerical optimization respectively. The paper also vividly presented a way to enhance a simple algorithm from a simple version to a powerful one.
Meng, Zhenyu, and Jeng-Shyang Pan. "A competitive QUasi-Affine TRansformation Evolutionary (C-QUATRE) algorithm for global optimization." Systems, Man, and Cybernetics (SMC), 2016 IEEE International Conference on. IEEE, 2016. (Paper/ Code) (Paper in Evolutionary Computation)
Meng, Zhenyu, and Jeng-Shyang Pan. "QUasi-affine TRansformation Evolutionary (QUATRE) algorithm: a parameter-reduced differential evolution algorithm for optimization problems." Evolutionary Computation (CEC), 2016 IEEE Congress on. IEEE, 2016. (Paper/ Code) (Paper in Evolutionary Computation)