文献类型：期刊文献

英文题名：An Adaptive Block-LSTM for Arcing Prediction in Pantograph-Catenary Systems

作者：Li, Te[1]; Fu, Jiaojiao[1]; Guo, Yi[1]; Liang, Yuhao[1]

机构：[1] East China University of Science and Technology, School of Information Science and Engineering, Shanghai, China

年份：2026

期号：2026

起止页码：20

外文期刊名：Proceedings of the International Conference on Computer Supported Cooperative Work in Design, CSCWD

收录：EI(收录号：20262921130539)

语种：英文

外文关键词：Computation theory - Data accuracy - Forecasting - Long short-term memory - Pantographs - Risk management - Safety factor - Subways

摘要：Arcing-induced high temperatures are a key factor in pantograph-catenary wear, yet real-world operational data is largely inaccessible due to strict rail transit safety constraints. Moreover, the original industrial dataset exhibits severe distribution imbalance, with extreme arcing events being rare yet highly consequential. To overcome both the unavailability and sparsity of real data, the underlying distribution of key system parameters is first estimated using kernel density estimation (KDE), and a realistic synthetic dataset is generated through a Conditional Tabular Generative Adversarial Network (CTGAN), which preserves the statistical characteristics of the original data. Building on this foundation, an Adaptive Block-Long Short-Term Memory (LSTM) network is proposed to accurately predict arcing temperatures in subway systems. This model leverages LSTM units to capture temporal dynamics, while block-wise partitioning guided by local data density effectively mitigates distribution heterogeneity and enhances prediction accuracy in underrepresented but high-risk regions. Furthermore, the Block-LSTM architecture is optimized using a genetic algorithm to improve predictive performance. Experimental results demonstrate that the proposed approach achieves high forecasting accuracy and consistently outperforms baseline models in terms of stability, generalization, and robustness to data sparsity. ? 2026 IEEE.