Multiobjective Evolutionary Multitasking With Two-Stage Adaptive Knowledge Transfer Based on Population Distribution

Multi-tasking optimization can usually achieve better performance than traditional single-tasking through knowledge transfer between tasks. However, current multi-tasking algorithms have some deficiencies. For high similarity problems, the that accelerate convergence rate of tasks has not been fully taken advantages of. low probability generating negative is high, which may result in degradation. In addition, methods proposed previously do consider how to deal with situation population falls into local optimum. To solve these issues, a two-stage adaptive evolutionary algorithm based on distribution, labeled as EMT-PD, proposed. EMT-PD and improve extracted from model reflects search trend whole population. At first stage, an weight used adjust step size individual's search, reduce impact transfer. second stage transfer, range further adjusted dynamically, diversity be beneficial for jumping out Experimental results multi-objective test suites show superior other six state-of-the-art multi/single-tasking algorithms. investigate effectiveness many-objective suite also designed this paper. The experimental new demonstrate competitiveness EMT-PD.