Evolutionary Multitask Optimization with Multi-Element Transfer

Evolutionary multitask optimization (EMTO) aims to utilize useful information between tasks and optimize multiple tasks at the same time. This is an emerging research direction in evolutionary computation. In recent years, many studies have promoted the progress of EMTO. Nevertheless, most algorithms are only designed for only one similar situation or choose a single transfer strategy in the algorithm, which limits the performance of the algorithm in problems with multiple similar situations. To solve this problem, this paper proposes an algorithm that transfers three evolutionary elements, i.e., the direction, distance, and position, via differential vectors, evolutionary paths, and the distribution of optimal solutions, respectively. Generally, these three elements control the generation of offspring in most cases of evolutionary optimization. Therefore, the combination of these three evolutionary elements is expected to solve multitask optimization problems with multiple similar situations more efficiently. At the same time, this paper designs an adaptive strategy to control the number of transferred offspring for each transfer strategy to avoid negative transfer. In order to verify the effectiveness of the proposed algorithm, comparisons on a single-objective multitask benchmark with classic and recently advanced algorithms are conducted, and the results verify the effectiveness of the designed algorithm.