Integrated Framework for Feature Selection and Dynamic Performance Prediction of Overhead Conductor Rail

Overhead conductor rail (OCR) is essential for traction power supply. However, as operating speeds increase, the dynamic performance of the pantograph-overhead conductor rail system (POCR) deteriorates significantly. To achieve low-cost and efficient monitoring, a sparse autoencoder (SAE) is introduced to evaluate node importance based on the encoding weight matrix. By incorporating a minimum spacing constraint, the master nodes contributing most significantly to the dynamic performance of the POCR are identified and found to be predominantly distributed at the supports. Subsequently, a bidirectional long short-term memory (BiLSTM) model is developed to predict the contact force and contact point movement using the vertical displacements of these master nodes as inputs. Results indicate that the root mean square error for contact force and contact point movement are 0.096 and 0.087, respectively, demonstrating the accuracy of the proposed method in capturing non-linear dynamic responses of the POCR.