2
Single objective optimization algorithms are the foundation upon which more complex methods, like
multi-objective, niching, and constrained optimization algorithms are built. Consequently,
improvements to single objective optimization algorithms are important because they can impact other
domains as well. These algorithmic improvements depend in part on feedback from trials conducted
with single objective benchmark functions, which themselves are the elemental building blocks for
more complex tasks, like dynamic, niching, composition, and computationally expensive problems. As
algorithms improve, ever more challenging functions must be developed. This interplay between
methods and problems drives progress, so we have developed the 2022 Special Session on Real-
Parameter Optimization to promote this symbiosis.
Improved methods and problems sometimes require updating traditional testing criteria. In recent
years, many novel optimization algorithms have been proposed to solve the bound-constrained, single
objective problems offered in the CEC’05
[1]
, CEC’13
[2]
, CEC’14
[3]
, CEC’17
[4]
, CEC’20
[4]
, and CEC’21
[6]
special sessions on Real-Parameter Optimization. Considering the comments on the CEC’20 test suite,
we organized this competition on real parameter single objective optimization.
Participants are required to send their results to the organizers in the format specified in this technical
report. Based on these results, organizers will present a comparative analysis that includes statistical
tests on convergence performance to compare algorithms with similar final solutions.
Participants may not explicitly use the equations of the test functions, e.g. to compute gradients. This
competition also excludes surrogate and meta-models. Papers on novel concepts that help us to
understand problem characteristics are also welcome. C, Python, and MATLAB codes for 2022 test
suite can be downloaded from the website below:
https://github.com/P-N-Suganthan
1. Introduction to the 2022 Benchmark Suite
1.1. Some Definitions:
All test functions are minimization problems defined as follows:
: number of dimensions.
: the shifted global optimum (defined in “shift_data_x.txt”), which is randomly
distributed in
. All test functions are shifted to and are scalable.
Search range:
. For convenience, the same search ranges are defined for all test functions.
: rotation matrix. Different rotation matrix is assigned to each function and each basic function.
Considering that linkages seldom exist among all variables in real-world problems, 2022 randomly
divides variables into subcomponents. The rotation matrix for each set of subcomponents is generated
from standard normally distributed entries by Gram-Schmidt ortho-normalization with condition
number c that is equal to 1 or 2.
