Causal-Guided Individual-Level Evolutionary Multitask Optimization: A Case Study on Water Bath Drawing

Evolutionary multitask optimization aims to exploit intertask relevant information to efficiently solve multiple tasks simultaneously. However, existing algorithms primarily perform intertask knowledge transfer at the population level and do not quantitatively evaluate the potential causal effect of each individual. Consequently, individual-level transfers are not sufficiently selective, while intratask knowledge transfer is often neglected. To address these issues, we propose a causal-guided individual-level evolutionary multitask optimization algorithm. During initialization, random samples are generated from the decision space of the source task. Small perturbations are applied to the samples, which are then mapped to a unified search space to create the random samples for the target task. Within this space, samples from different tasks are paired based on Euclidean distance. For each task-specific paired sample set, the samples are divided into treatment and control groups according to whether they undergo transfer, and causal forest models are trained separately for each task. During evolution, this strategy is periodically activated for fine-grained individual-level knowledge transfer. Meanwhile, an elite-guided intratask knowledge transfer strategy uses the best individual in each generation to enhance local search. Experimental results on multiple benchmark problems and a case study on optimizing both liquid-level and concentration controller parameters for water bath drawing demonstrate that the proposed algorithm outperforms state-of-the-art approaches.