Autoencoder evolutionary algorithm for large-scale multi-objective optimization problem

Multi-objective optimization problems characterized by a substantial number of decision variables, which are also called large-scale multi-objective optimization problems (LSMOPs), are becoming increasingly prevalent. Traditional evolutionary algorithms may deteriorate drastically when tackling a large number of decision variables. For LSMOPs, the dimensionality of the decision variables needs to be reduced and the algorithm needs to be designed according to the characteristics of divide-and-conquer. The autoencoder evolutionary algorithm (AEEA) is proposed based on autoencoder dimensionality reduction, the grouping of decision variables, and the application of divide-and-conquer strategies. The proposed algorithm is compared with other classical algorithms. The experiment result shows that AEEA achieves excellent convergence and diversity, and still performs well in decision variables of higher dimensions. Finally, it is verified that the autoencoder improves the running time of the proposed algorithm.