Scholarlink

The transmission shaft system, as a critical component of ship power transmission, its stability is directly related to the safety of ship navigation. Hence, to accurately predict the fault features of the transmission shaft system is of paramount importance. However, Long Short-Term Memory (LSTM) prediction model typically uses static parameter configurations, which limits its ability to adapt to the dynamic changes in fault features. Therefore, this paper introduces an adjustable factor ηt for LSTM and proposes an adaptive adjustment mechanism based on Belief Rule Base (BRB). By analyzing the statistical characteristics of LSTM input sequence in real-time, the adjustable factor ηt is dynamically inferred, thereby enhancing the predictive performance. Furthermore, the predicted fault feature sequences are inputted into Support Vector Machine (SVM) model to achieve the prediction of fault types. Finally, the prediction experiments of the unbalanced fault feature sequences in the transmission shaft experimental platform are implemented to show that the proposed method has significant improvement in prediction accuracy compared to the traditional LSTM model.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LSTM Time Series Prediction Model with Adjustable Inputs and Application in Fault Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Juntao Shi,
  • Xiaobin Xu,
  • Haohao Guo,
  • Jun Dong,
  • Jie Sun,
  • Hua Mao,
  • Hailun Qiu

摘要

The transmission shaft system, as a critical component of ship power transmission, its stability is directly related to the safety of ship navigation. Hence, to accurately predict the fault features of the transmission shaft system is of paramount importance. However, Long Short-Term Memory (LSTM) prediction model typically uses static parameter configurations, which limits its ability to adapt to the dynamic changes in fault features. Therefore, this paper introduces an adjustable factor ηt for LSTM and proposes an adaptive adjustment mechanism based on Belief Rule Base (BRB). By analyzing the statistical characteristics of LSTM input sequence in real-time, the adjustable factor ηt is dynamically inferred, thereby enhancing the predictive performance. Furthermore, the predicted fault feature sequences are inputted into Support Vector Machine (SVM) model to achieve the prediction of fault types. Finally, the prediction experiments of the unbalanced fault feature sequences in the transmission shaft experimental platform are implemented to show that the proposed method has significant improvement in prediction accuracy compared to the traditional LSTM model.