A surrogate-assisted multitask knowledge transfer optimization algorithm and application

doi:10.1016/j.knosys.2026.116576

A surrogate-assisted multitask knowledge transfer optimization algorithm and application

2026-07-01

0

PRE

AI

DOI：10.1016/j.knosys.2026.116576

Original

Original request for help

Save

Abstract

En 中文

Conventional optimization methods, which initiate the search from scratch without leveraging prior knowledge, are always time-consuming and resource-inefficient. However, most problems in the real world do not exist in isolation, such as those in the fields of aerospace, transportation, etc. Abundant data and knowledge of similar optimized tasks can be utilized by transfer optimization to help the optimization of target task, which has attracted growing attention in recent years. However, when the similarity between two tasks is insufficient, the transferred knowledge is likely to mislead the search, which is called negative transfer. To address this issue, a surrogate-assisted multitask knowledge transfer optimization algorithm (SATO) is proposed in this article. In SATO, an intermediate task is constructed to enhance the similarity among multiple tasks, thereby facilitating more effective transfer optimization. In addition, a distance-based source database construction mechanism is devised. Through this mechanism, the source database consists of samples most relevant to the target task, rather than those with superior performance but low similarity. This ensures that the transfer optimization process is supported by a more relevant and informative source database. Furthermore, a transfer integration model is established among the source, intermediate, and target tasks in SATO. This model facilitates the selection of new promising samples and continuously refine the predictive model for the target task, ultimately achieving efficient transfer optimization based on multitask knowledge. Experimental results demonstrate that the proposed SATO exhibits superior performance when compared with seven algorithms, achieving the best results in 53.9% out of the benchmark cases. Eventually, SATO is applied to the shape optimization of unmanned underwater vehicle (UUV). The results further validate its competitiveness in handling computationally expensive engineering problems.

Keywords:

Global optimization

Transfer optimization

Unmanned underwater vehicle

Multitask knowledge

A surrogate-assisted multitask knowledge transfer optimization algorithm and application

Abstract

Journal

Organization

Cited Papers