Bayesian optimization bidirectional LSTM approach for the condition assessment of underground-operating trains
Yu-Ling Wang, Yuhang Lu, Yanke Tan, Wai Kei Ao, Yi-Qing Ni, Qing-Chen Tang
Abstract
Abstract Underground railways play a crucial role in global intercity transport networks, necessitating the implementation of diverse measures to mitigate vibrations during train operations. However, with the variable damping, the structures of underground trains can inadvertently impact passenger’s comfort when taking them. Consequently, the development of the online monitoring system becomes imperative to assess the operational conditions of these trains. This research applies the ISO2631 standard to analyze the dynamic responses of train’s accelerations, utilizes the ride comfort index to determine the operational state of the train, and uses online monitoring data to evaluate its overall conditions. The study proposes an online monitoring system that utilizes the long short-term memory (LSTM) algorithm, which has demonstrated effectiveness in time-series prediction and identification tasks. By learning from historical and future signal segments, the LSTM algorithm enables the diagnosis and identification of underground train-operating conditions under varying working conditions. To enhance the accuracy of prediction results, the algorithm is optimized by adopting the bi-directional structure and Bayesian optimization method. Quantitative analyses demonstrate that the optimized bi-directional LSTM model achieves a correlation up to 94.32% for overall dataset and 90.45% on test dataset. Finally, an illustrative case is presented to highlight the performance of the proposed method in assessing the conditions of underground trains.