.png){kind=icon}
.png){kind=icon}
Return
Distributed Data-Driven Evolutionary Optimization With Surrogate Ensemble Strategies
.png){kind=icon}
PRE
AI
DOI：10.1109/tetci.2026.3683108.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Surrogate-assisted evolutionary algorithms (SAEAs) have emerged as effective solutions for addressing data-driven optimization problems (DOPs). However, the advent of distributed DOPs, attributed to the widespread existence of distributed data scenarios, has promoted the development of distributed SAEAs. Distributed SAEAs encounter the challenge of disparate nodes independently constructing local models based on their observed data, complicating evolutionary coordination. To address this challenge, we propose a novel surrogate model ensemble framework, which aims to effectively guide the evolution across distributed nodes and enhance the performance of distributed SAEAs. The framework consists of two main components, an Adaptive boosting Surrogate training strategy (AdaSurrogate, AS) and an Adaptive model Aggregation strategy (AdaAggregation, AA). AS dynamically adjusts the sample weights of surrogate models on edge clients, aiming to enhance the accuracy of local surrogate models. AA aggregates different client models across rounds, aiming to improve the accuracy of the global surrogate model and its adaptability to distributed data heterogeneity. AS and AA are applied to improve the Edge-Cloud Co-Evolutionary Algorithm (ECCoEA), an existing advanced distributed SAEA, resulting in ECCoEA-ASAA. Experiments show that ECCoEA-ASAA not only exhibits a huge improvement over the original ECCoEA in both independent and identically distributed (i.i.d.) and non-i.i.d. scenarios, but also outperforms some state-of-the-art SAEAs in non-i.i.d. scenarios. Additionally, an ablation study is also conducted to demonstrate and analyze the effectiveness of AS and AA in enhancing the performance of distributed SAEAs.
Keywords:
Data-driven optimization
distributed optimization
surrogate-assisted evolutionary algorithm
ensemble learning
Journal
I
IF:
6.5
Papers:
1.4K
Citations:
4.5K