A Many-Objective Optimization Evolutionary Algorithm Based on Double Surrogate-Assisted Adaptive Guiding Evolutionary Direction

Abstract In expensive Many-objective Optimization Problems, Surrogate-Assisted Evolutionary Algorithms (SAEAs) are often used to reduce the number of original Function Evaluations (FE). However, traditional SAEAs may suffer from long running times due to the complex computational demands of the surrogate models themselves. In this paper, we propose a novel algorithm called " A Many-Objective Optimization Evolutionary Algorithm Based on Double Surrogate-Assisted Adaptive Guiding Evolutionary Direction (DSAG)", aimed at further reducing the running time of SAEAs. The proposed algorithm employs two surrogate models: one for predicting the diversity of solutions and another for predicting the convergence value of solutions, and then sorts the solutions based on the prediction results. Afterwards, the algorithm adaptively adjusts the type and number of candidate solutions according to the "Convergence-Related Average Move Distance (CAMD)" proposed in this paper. This algorithm can adaptively bias towards either the convergence exploration stage or the diversity exploration stage. Compared with other classical surrogate-assisted evolutionary algorithms, the proposed algorithm first categorizes the decision variables and uses the categorized data to train two surrogate models, which makes the overall complexity and running time of the algorithm superior to others. Finally, we compared the performance of this algorithm on different problems using benchmark test functions to verify its ability to explore diversity and improve convergence. Then, we compared the running time of the algorithm with changes in the target dimension, decision variable dimension, and number of Original FE on the same test problems, and demonstrated the superior performance of the algorithm.