Evolutionary computation-based multitask learning network for railway passenger comfort evaluation from EEG signals

As a core indicator of the railway service industry, the comfort of the railway passengers is an important aspect to measure the advancement of railway technology. The accurate evaluation of railway passenger comfort has received much attention from academia and industry. The EEG-based comfort evaluation method has been hailed as the gold standard due to its rich quantity of information and objectivity. In this study, an evolutionary computation-based multitask learning network named EEG-DEMTL is proposed to evaluate railway passenger comfort from EEG signals. In this network, the multitask learning structure fully exploits the relationship between subtasks such as passenger emotion and the main task, which is passenger overall comfort. Furthermore, the weight definition method based on the differential evolution (DE) algorithm is used to select the best weight setting. To verify the validity of our proposed method, field experiments in a high-speed railway (HSR) are designed, and comfort perceptions and EEG signals of 20 passengers are collected. Compared with the baseline models, which include the support vector machine, K-nearest neighbour and decision trees, the proposed EEG-DEMTL model achieves the best performance. In addition, the comparison results between several weight settings of MTL show that the DE weights can improve the evaluation performance by 6.30%. This research proposes a novel EEG-based method to meet the requirements of railway passenger comfort evaluation and offers a neurological explanation for railway passenger comfort. © 2023 Elsevier B.V.