• Danial Yazdani

  Faculty of Engineering & Information Technology, University of Technology Sydney, Ultimo 2007, Australia
  Email: danial.yazdani@gmail.com

• Michalis Mavrovouniotis

  ERATOSTHENES Centre of Excellence, Limassol, Cyprus
  Email: michalis.mavrovouniotis@eratosthenes.org.cy

• Changhe Li

  the School of Automation, China University of Geosciences, Wuhan 430074, China
  Email: changhe.lw@gmail.com

• Wenjian Luo

  Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies, School of Computer Science and Technology, Harbin Institute of Technology and Peng Cheng Laboratory, Shenzhen 518055, China
  Email: luowenjian@hit.edu.cn

• Mohammad Nabi Omidvar

  School of Computing, University of Leeds, and Leeds University Business School, Leeds LS2 9JT, United Kingdom
  Email: m.n.omidvar@leeds.ac.uk

• Amir H. Gandomi

  Faculty of Engineering & Information Technology, University of Technology Sydney, Ultimo 2007, Australia and University Research and Innovation Center (EKIK), Obuda University, Budapest 1034, Hungary
  Email: Gandomi@uts.edu.au

• Trung Thanh Nguyen

  Liverpool Logistics, Offshore and Marine (LOOM) Research Institute, Faculty of Engineering and Technology, School of Engineering, Liverpool John Moores University, Liverpool L3 3AF, United Kingdom
  Email: t.t.nguyen@ljmu.ac.uk

• Jurgen Branke

  Operational Research and Management Sciences Group at Warwick Business school, University of Warwick, Coventry CV4 7AL, United Kingdom
  Email: Juergen.Branke@wbs.ac.uk

• Xiaodong Li

  School of Science (Computer Science), RMIT University, Melbourne, 3001, Australia
  Email: xiaodong.li@rmit.edu.au

• Shengxiang Yang

  Center for Computational Intelligence (CCI), School of Computer Science and Informatics, De Montfort University, Leicester LE1 9BH, United Kingdom
  Email: syang@dmu.ac.uk

• Xin Yao

  Research Institute of Trustworthy Autonomous Systems (RITAS), and Guangdong Provincial Key Laboratory of Brain inspired Intelligent Computation, Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, China. He is also with the CERCIA, School of Computer Science, Birmingham B15 2TT, United Kingdom
  Email: xiny@sustech.edu.cn

Search space of many real-world optimization problems is dynamic in terms of the objective function, the number of variables, and/or constraints. Optimization problems that change over time and need to be solved online by an optimization method are referred to as dynamic optimization problems (DOPs). To solve DOPs, algorithms not only need to find desirable solutions but also be able to react to the environmental changes in a timely manner and find a new solution when the previous one becomes suboptimal. In this competition, we focus on evolutionary dynamic optimization methods (EDOs) for unconstrained single-objective continuous DOPs. EDOs are important class of algorithms as they form the foundation for more complex methods, such as those used in robust optimization over time, as well as large-scale and constrained dynamic optimization. This competition aims at providing a common platform for fair and easy comparison of EDOs. The competition allows competitors to run their own EDO on the problem instances generated by the generalized moving peaks benchmark (GMPB) with different characteristics and levels of difficulty.

The landscapes generated by GMPB are constructed by assembling several promising regions with a variety of controllable characteristics ranging from unimodal to highly multimodal, symmetric to highly asymmetric, smooth to highly irregular, and various degrees of variable interaction and ill-conditioning. An example of a 2-dimensional landscape generated by GMPB is shown in Figure 1.

12 different problem instances with different characteristics are used in this competition. To generate these problem instances, the participants need to set four parameters PeakNumber, ChangeFrequency, Dimension, and ShiftSeverity in “main.m” according to the values shown in Table 1.

Offline error is used as the performance indicator in this competition :

`E_(o)=1/(Tϑ)sum_(t=1)^Tsum_(c=1)^ϑ(f^"(t)"(vecx^(∘"(t)"))-f^"(t)"(vecx^(**"("(t-1)ϑ+c")")))`

where  `vecx^(∘"(t)")` is the global optimum position at the tth environment, T is the number of environments, 𝜗 is the change frequency, c is the fitness evaluation counter for each environment, and  `vecx^(**((t-1)ϑ+c))` is the best found position at the cth fitness evaluation in the tth environment. Offline error is the average of current error values over optimization process. In the source code, in each run, the values of the current errors are stored in “Problem.CurrentError” at the end of each run, the offline error is obtained.

1. Participants are NOT allowed to change the random seed generators in the code.
2. Participants are NOT allowed to modify “BenchmarkGenerator.m” and “fitness.m” files in the source code.
3. Participants are NOT allowed to tune the parameters of their algorithm for individual problem instance. In other words, the values of the parameters of the algorithm must be the same for solving all problem instances.
4. The problem instances must be considered as complete blackboxes and the algorithms must NOT use any of the internal parameters of GMPB.
5. The algorithm can be informed about the environmental changes. Consequently, participants do not need to use any change detection mechanism in their algorithms. Note that it is also accepted if a change detection component is used.
6. Each participant can submit more than one algorithm.
7. Competitors can participate with either newly proposed algorithms (unpublished) or their previously published algorithms.
8. Winners will be required to share their algorithm's source code for result verification. This code will remain confidential and will not be published.

For each algorithm, the competitors must provide a compressed folder (named as the algorithm's name) which contains 13 files. The first file is a document containing the following information :

1. Title
2. Names, affiliations, and emails of participants
3. A short description of the algorithm, including the used optimizer (e.g., PSO or DE), how the population is managed and controlled (e.g., bi-population, multi-population with fixed number of sub-populations, or multi-population with clustering), and whether explicit memory (archive) is used or not.
4. The following table which shows the best, worst, average, median, and standard deviation of the offline error values obtained by 31 runs for each problem instance.

In addition, this folder also must contain 12 text files where each file contains the 31 offline error values obtained by the algorithm in 31 runs for each problem instance. Each text file must be named according to the corresponding problem instance (e.g., “F10.dat” which contains 31 entries obtained by solving F10).

References

[1] D. Yazdani, M. N. Omidvar, R. Cheng, J. Branke, T. T. Nguyen, and X. Yao, “Benchmarking continuous dynamic optimization: Survey and generalized test suite,” IEEE Transactions on Cybernetics, vol. 52(5), pp. 3380-3393, 2020.

[2] M. Peng, Z. She, D. Yazdani, D. Yazdani, W. Luo, C. Li, J. Branke, T. T. Nguyen, A. H. Gandomi, Y. Jin, and X. Yao, “Evolutionary dynamic optimization laboratory: A matlab optimization platform for education and experimentation in dynamic environments,” arXiv preprint arXiv:2308.12644, 2023.