Ten New Benchmarks for Optimization

Xin-She Yang

DOI: https://doi.org/10.1007/978-981-99-3970-1_2

2023-08-30

Abstract:Benchmarks are used for testing new optimization algorithms and their variants to evaluate their performance. Most existing benchmarks are smooth functions. This chapter introduces ten new benchmarks with different properties, including noise, discontinuity, parameter estimation and unknown paths.

Optimization and Control,Artificial Intelligence,Neural and Evolutionary Computing

What problem does this paper attempt to address?

The paper aims to address the limitations of existing benchmarks in optimization algorithm testing. Specifically, most current benchmark test functions are smooth functions with a limited number of optimal solutions, and their search domains are usually regular with simple boundaries or constraints. However, these benchmarks are far from the optimization problems encountered in real-world applications, and thus their test results may not accurately reflect the performance of algorithms on actual problems.

To address this issue, the authors propose 10 new benchmark test functions with different characteristics, including noise, discontinuities, non-differentiability, multi-layer discrete values, and optimal paths from variational methods. The design of these new benchmark test functions aims to better simulate the complex situations in real-world applications, thereby providing a more comprehensive evaluation of the performance of optimization algorithms.

By introducing these new benchmarks, the paper hopes to provide more challenges during algorithm validation, thereby encouraging researchers to develop more robust and practical optimization algorithms.