.png){kind=icon}
.png){kind=icon}
Return
Data-Driven Evolutionary Computation Under Continuously Streaming Environments: A Drift-Aware Approach
.png){kind=icon}
PRE
AI
DOI：10.1109/TEVC.2025.3589643.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Streaming data-driven evolutionary algorithms (SDDEAs) have emerged as a crucial paradigm in the area of data-driven optimization. However, current methods face critical limitations when handling unpredictable concept drifts in continuously evolving environments. To address this research gap, we propose DASE, a <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">D</u>rift-<underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A</u>ware <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">S</u>treaming <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">E</u>volutionary algorithm that features two key innovations. First, we introduce a hierarchical confidence drift detector that operates on a moving window over continuous data streams, identifying concept drifts by evaluating statistical deviations in model accuracy. Second, we propose a context-aware warm start mechanism that adaptively transfers knowledge from historical environments to the new environment using environmental similarity-based weighting. These dual innovations not only enables automatic segmentation of streaming data into coherence environments but also enhances optimization performance with the real-time responsiveness. Experimental evaluations on benchmark problems demonstrate that DASE significantly outperforms state-of-the-art algorithms across various drift scenarios, establishing it as a powerful method for addressing challenges in continuously streaming environment.
Keywords:
Concept drift
dynamic optimization
streaming data
streaming data-driven evolutionary algorithm (SDDEA)
surrogate model
Journal
IF:
12
Papers:
1.8K
Citations:
2.4W