Research on Multi-Objective Multitasking Algorithm Based on Dimensional Importance Assessment and Knowledge Transfer

In the field of multi-objective multi-task optimization, knowledge transfer is crucial to improve the performance of algorithms. However, most existing studies directly transfer individuals, which is prone to introduce irrelevant genes and cause negative transfer phenomena. For this reason, this paper proposes a multi-objective multitasking algorithm based on dimensional importance assessment and knowledge transfer. First, the algorithm designs indicators to assess the importance of dimensions. It screens out the key dimensions of the source and target tasks by analyzing the difference between the decision vectors of the optimal and the worst solutions in the population, combined with the out-of-bag error calculation method of random forest. Then, dimension matching is performed using Euclidean distance to ensure the correspondence of key dimensions of source and target tasks. Subsequently, non-dominated solutions are selected from the source task to transfer to the target task and verify the performance. For the successfully transferred solutions, the scaling relationship of the key dimensions is calculated, and they are transferred to the target task for adjustment according to the dimension-matching relationship. Experiments show that this method can effectively improve the knowledge transfer effect and reduce negative transfer.