A neural network guided dual-space search evolutionary algorithm for large scale multi-objective optimization

The curse of dimensionality caused by the increase of decision variables in large-scale multi-objective problems (LSMOPs) is still the current challenge. Although existing algorithms can simple large-scale multi-objective optimization problems. Nevertheless, a single search strategy might impact the solution of large-scale multiobjective optimization problems. To solve this problem, a dual-space search evolutionary algorithm for largescale multi-objective optimization is proposed. Firstly, in the decision space, a neural network assisted operator with adaptive strategy is introduced. Specifically, when the number of non-dominated solutions is decreasing, the neural network is adopted to optimize the solutions with poor fitness for breaking away from local optimality. After that, the objective space of population is divided into several sub-regions by k-means clustering strategy. The solutions in these subregions are mapped onto the decision space through the inverse model, so that population can obtain as many non-dominated solutions as possible. Finally, the proposed algorithm is tested on a real-life problem which is Time-varying Ratio Error Estimation (TREE) and two benchmark suites which are large-scale multi-objective optimization problem (LSMOP) and unconstrained front (UF). The results show that the proposed algorithm exhibits competitive performance compared to other state-of-the-art algorithms on Inverted Generational Distance (IGD) Indicator and Hyper-volume (HV) Indicator.