Optimal operator selection based on the hybrid operator selection strategy

At present, the multiobjective optimization algorithm based on decomposition (MOEA/D) has become a typical algorithm for solving multiobjective problems. Following extensive experiments, researchers found that the choice of mutation operators in MOEA/D has a great influence on the quality of the solutions, and a suitable operator can largely improve the algorithm's performance. This paper proposes a hybrid operator selection strategy based on feedback and support vector machine (SVM) classification, and applies it to the framework of MOEA/D. The algorithm divides the whole evolutionary process into several stages, and each stage is divided into two parts. In the first part, multiple mutation operators are given the same computing resources to generate offspring, then, we test the performance of each operator and elect the operator with the best performance. Meanwhile, an SVM classifier is trained by taking the truly evaluated offspring as the training set. Good operators do not always generate good offspring. Therefore, in the second part, we use the best performing operator selected in the first part to generate offspring, while using the trained classifier to filter all the newly generated solutions. The real evaluation should be carried out only when the offspring are identified as positive. Otherwise, it should be ignored to save the real evaluation computing resources. Additionally, the number of all positive individuals is recorded in the second part, which is when we decide whether to continue using this operator at the next stage or to reselect a more suitable operator. Experimental results verify the effectiveness of this strategy.

Automated summary

This paper proposes a hybrid operator selection strategy based on feedback and support vector machine (SVM) classification for improving the multiobjective optimization algorithm. Experimental results verify the effectiveness of this strategy.