Data-Driven Optimization Based on Random Forest Surrogate
Yuan Guang Zheng, Xiaogang Fu, Yanwen Xuan · 2019
For the practical problem that there are no exact evaluation functions to evaluate candidate solutions, it can only be solved by data-driven method. To solve these problems, Random Forests (RF) is used as surrogates to approximate the evaluation functions and constraint functions of constrained combinatorial optimization problems. In addition, support vector machine (SVM) model are introduced to rectify the surrogate-assisted fitness evaluations. The proposed error correction strategy is tested by a series of benchmarks of MOPKs. The results show that the proposed error correction strategy is more effective when RF is used as the surrogates to solve multi-objective combinatorial optimization problems with constraints.