Addressing expensive many-objective optimization problems (MaOPs) is a formidable challenge owing to their intricate objective spaces and high computational demands. Surrogate-assisted evolutionary algorithms (SAEAs) have gained prominence because of their ability to tackle MaOPs efficiently. They achieve this by using surrogate models to approximate objective functions, significantly reducing their reliance on costly evaluations. However, the effectiveness of many SAEAs is hampered by their reliance on various surrogate models and optimization strategies, which often result in suboptimal prediction accuracy and optimization performance. This study introduces a novel approach: an activity level based surrogate-assisted reference vector guided evolutionary algorithm specifically designed for expensive MaOPs. Utilizing the Kriging model and an angle penalty distance criterion, this algorithm effectively filters solutions that require evaluation using the original function. It employs a fixed number of training sets,that are updated via a two-screening strategy that leverages activity levels to refine population screening. This process ensures that the reference vector progressively aligns more closely with the Pareto fronts,which is enhanced by the deployment of adjusted adaptive reference vectors, thereby improving the screening precision. The proposed algorithm was tested against six contemporary algorithms using the DTLZ, WFG, and MaF test suites.The experimental results show that the proposed method outperforms other algorithms in most problems. Furthermore, its application to the cloud computing task scheduling problem underscores its practical value, demonstrating its notable effectiveness. The experimental outcomes attest to the robust performance of the algorithm across both test scenarios and real-world applications.

An activity level based surrogate-assisted evolutionary algorithm for many-objective optimization