Benchmarking Continuous Dynamic Optimization: Survey and Generalized Test Suite.

Area of Science:

• Computational intelligence and optimization
• Swarm and evolutionary computation

Background:

• Real-world optimization problems often involve dynamic changes, requiring adaptive algorithms.
• Evaluating dynamic optimization algorithms necessitates robust benchmarks with controllable characteristics.
• Existing benchmarks have limitations in capturing diverse problem features and dynamic properties.

Purpose of the Study:

• To comprehensively review and identify shortcomings of current dynamic optimization benchmarks.
• To propose a highly configurable benchmark suite, the Generalized Moving Peaks Benchmark (GMPB).
• To assess the performance of established and novel dynamic optimization algorithms using the GMPB.

Main Methods:

• A systematic review of existing dynamic optimization benchmark suites.
• Development of the Generalized Moving Peaks Benchmark (GMPB) with configurable properties.
• Experimental evaluation of selected optimization algorithms on GMPB-generated problem instances.

Main Results:

• Existing benchmarks inadequately represent the complexity and dynamism of real-world optimization problems.
• The GMPB generates problem instances with diverse, controllable features (ill-conditioning, variable interactions, shape, complexity).
• Proposed GMPB components exhibit high dynamism in gradients, heights, optimum locations, and other properties.
• Well-known optimizers and dynamic algorithms demonstrated poor performance on GMPB-generated challenges.

Conclusions:

• The Generalized Moving Peaks Benchmark offers a more comprehensive and challenging testbed for dynamic optimization.
• Current algorithms struggle with the multifaceted dynamic challenges introduced by the GMPB.
• Further research is needed to develop more robust and adaptive algorithms for dynamic optimization problems.