SDDMO-Bench: New benchmark for streaming multi-objective optimization

SDDMO-Bench, a new benchmark suite by researchers Wenjie Xiao, Hui Bai, and Junhao Chen, targets streaming data-driven dynamic multi-objective optimization (DDMO). In these problems, algorithms must track time-varying Pareto fronts using only sequential observations under concept drift—where the underlying data distribution shifts over time. Real-world applications like dynamic pricing, supply chain management, and adaptive control systems face this challenge, but evaluation is notoriously difficult because real datasets often lack ground-truth optima, drift annotations, and controllable conditions.

The proposed suite transforms classical dynamic multi-objective test problems into streaming environments by combining three mechanisms: intrinsic objective-mapping evolution, controllable distributional drift, and sequential data revelation. By pairing five representative time-dependent base functions with six distributional drift patterns, SDDMO-Bench constructs 30 scenarios with varying levels of non-stationarity, problem complexity, and sample-distribution variation. Experiments with representative evolutionary algorithms show the suite delivers challenging and discriminative tests for ranking algorithm performance across adaptability, robustness, and Pareto-front tracking. The authors submitted the work to the IEEE MIND Conference, with the paper and supplementary materials available on arXiv.