Evolutionary Algorithms With Blind Fitness Evaluation for Solving Optimization Problems With Only Fuzzy Fitness Information

Evolutionary algorithms (EAs) show strong adaptability in solving the complex optimization problems. Fitness evaluation is an important step of EA. In this step, the fitness of individuals is calculated by fitness evaluation methods. However, there are many complex optimization problems in which the existing fitness evaluation methods cannot express solutions’ fitness by crisp values but only fuzzy fitness information. Even if some fitness evaluation methods can predict crisp values to express the individuals’ fitness, these methods must “see” some crisp fitness information in advance. These shortcomings of existing fitness evaluation methods make EAs unable to work effectively in many optimization tasks. In this article, we propose the concept of blind fitness evaluation. A blind fitness evaluation method (BFEM) can output individuals’ crisp fitness information without “seeing” any crisp fitness information in advance. Based on the concept, we propose a BFEM which uses an artificial neural network, named modified fitness comparison network (MFCN) to evaluate the fitness of individuals. The MFCN can be trained by fuzzy fitness information, and the trained MFCN can predict the crisp fitness information of any individual. By applying the BFEM to EAs, the EAs with blind fitness evaluation are obtained and the algorithms can effectively solve the complex optimization problems with only fuzzy fitness information. The proposed BFEM is compared with several popular fitness evaluation methods that rely on crisp fitness information. Experimental results show that the fitness evaluation accuracy of MFCN trained based on only fuzzy fitness information is similar to or higher than that of the fitness evaluation methods relying on crisp fitness information. The EAs with MFCN also show higher performance than the EAs with the popular fitness evaluation methods in search ability and solution quality.