Evolutionary Multitasking with Adaptive Tradeoff Selection Strategy

As a new emerging evolutionary framework, evolutionary multitasking aims to optimize multiple tasks simultaneously. Knowledge transfer is an important component of evolutionary multitasking. How to extract and transfer knowledge significantly affects the performance of the algorithm. A serious challenge for evolutionary multitasking is the inappropriate knowledge transfer or insufficient exploration and exploitation. To address this challenge, an evolutionary multitasking with adaptive tradeoff selection strategy (EMT-ATS) is proposed. To enhance global exploration and local exploitation during the evolution, an adaptive tradeoff selection mechanism is developed to select promising offspring during different stages to guide the population toward more promising solution regions. In addition, a Cohen’s d indicator-based is used to adjust knowledge transfer. To verify the effectiveness of the proposed EMT-ATS, a series of experiments are conducted with several popular evolutionary multitasking algorithms on multitasking benchmark problems. In addition, a multitask optimization problem involving two real-world problems is used to validate the practicability of the proposed EMT-ATS. Experimental results demonstrate the effectiveness of the proposed EMT-ATS.