Multi-objective particle swarm optimization for selecting optimal grid learning without replacement sampling

A novel multi-objective particle swarm optimization is proposed in this paper, which can select the optimal grid learning without replacement sampling (WRSMOPSO). The algorithm first adaptively groups the population according to the optimal number of grids, then selects the optimal grid for each group without replacement sampling, and finally selects the optimal solution randomly from the selected optimal grids for learning. In order to ensure the diversity and convergence of the algorithm, this paper also established a new external archive control strategy based on grid technology. The WRSMOPSO and the classic MOPSO are simulated experiments on the benchmark function. Experimental results show that the WRSMOPSO can effectively improve the inverted generational distance (IGD) and hypervolume indicator (HV), which compares with classic MOPSO, the WRSMOPSO shows good overall performance.