Surrogate-assisted Evolutionary Multitasking Optimization based on Transfer Security

In practical engineering design processes, optimization tasks are often conducted in a staged and distributed manner, where different design phases may exhibit underlying structural correlations. The intermediate models and partial solutions derived from such processes can offer valuable prior knowledge for the current optimization task, thereby enhancing the overall optimization efficiency. However, in distributed collaborative scenarios, existing studies frequently overlook the potential risk of privacy leakage during the transfer process, which may lead to the exposure of intellectual property and weaken the engineering applicability of the algorithms. To address this issue, this paper proposes a Transfer Security-based Surrogate-assisted Evolutionary Multitasking Optimization Algorithm (TSMTEA). TSMTEA introduces a dual-layer knowledge transfer mechanism at both the model and genetic levels to fully exploit the complementary information across tasks. For model transfer, an adaptive selection and stacking-based model ensemble strategy is designed to effectively identify and filter out potential malicious models. In addition, the Advanced Encryption Standard algorithm is incorporated to ensure secure model transmission between tasks. For genetic transfer, a secure differential evolution procedure based on homomorphic encryption is constructed to enable inter-task genetic operations without revealing original genetic information. To evaluate the performance of the proposed method, systematic experiments are conducted on multiple synthetic benchmark suites as well as the neural architecture search problem. Experimental results demonstrate that the proposed algorithm achieves significant advantages in both convergence performance and privacy preservation.