An Improved Fuzzy Classifier-Based Evolutionary Algorithm for Expensive Multiobjective Optimization Problems with Complicated Pareto Sets.

Various surrogate-based multiobjective evolutionary algori-thms (MOEAs) have been proposed to solve expensive multiobjective optimization problems (MOPs). However, these algorithms are usually examined on test suites with unrealistically simple Pareto sets (e.g., ZDT and DTLZ test suites). Real-world MOPs usually have complicated Pareto sets, such as a vehicle dynamic design problem and a power plant design optimization problem. Such MOPs are challenging to construct reliable surrogates for surrogate-based MOEAs. Constructed surrogates with low accuracy are likely to make incorrect predictions and even mislead the search direction. In this paper, we propose an improved fuzzy classifier-based MOEA by leveraging the accuracy information of the classifier. The proposed algorithm is compared with five state-of-the-art algorithms on two well-known test suites with complicated Pareto sets and four real-world problems. Experimental results demonstrate the effectiveness of the proposed algorithm in solving realistic MOPs with complicated Pareto sets when only a limited number of function evaluations are available.