海洋工程装备与技术

海洋工程装备与技术 ›› 2024, Vol. 11 ›› Issue (4): 1-7.doi: 10.12087/oeet.2095-7297.2024.04.01

• •    下一篇

基于LSTM重构环境特征的异常值检测

周 雷1，苏 馨2，张 崎2*，黄 一2   

  1. 1. 海洋石油工程股份有限公司，天津，300451； 2. 大连理工大学船舶工程学院，辽宁 大连，116086
  • 出版日期:2025-02-21 发布日期:2025-02-21
  • 作者简介:周雷（1978— ），男，博士，高级工程师，研究方向：海洋工程设施完整性。
  • 基金资助:
    海洋工程数字孪生机理种子库及导管架结构、浮式设施数字孪生技术(Z2023SYENT1390)

Anomaly Detection of Environmental Features Based on LSTM-Reconstructed Data

ZHOU Lei1，SU Xin2，ZHANG Qi2*，HUANG Yi2   

  1. 1. Offshore Oil Engineering Co.， Ltd.， Tianjin 300451， China； 2. School of Naval Architecture and Ocean Engineering， Dalian University of Technology， Dalian 116086， Liaoning China
  • Online:2025-02-21 Published:2025-02-21

PDF (PC)

33

摘要： 随着环境监测技术的发展，准确识别数据中的异常值成为一个重要挑战。本研究提出了一种结合长短时记忆网络（LSTM）和随机森林模型的方法，用于预测和重构环境特征数据，进而实现对异常值的有效监测。首先，使用LSTM模型对环境特征如风速、风向等进行时间序列预测，然后以这些预测结果作为输入，应用随机森林模型对轴力进行预测。研究表明，通过对特征的重构，相较于直接的异常值监测方法，可以显著提高轴力预测的准确性。重构特征后数据的R2值、MAE（平均绝对误差）和RMSE（均方根误差）均优于原始特征数据。特别是R2值，由0.921提升至0.956，证明了模型在数据拟合上的显著提升。

关键词: 长短时记忆网络, 随机森林, 异常值监测, 特征重构

Abstract: With the advancement of environmental monitoring technologies， accurately identifying anomalies in data has become a crucial challenge. This study introduces a method combining Long Short-Term Memory (LSTM) networks and Random Forest models for predicting and reconstructing environmental feature data， thereby enabling effective anomaly monitoring. Initially， the LSTM model is used to forecast time series data for environmental features such as wind speed and direction. Subsequently， these forecasts are used as inputs to apply a Random Forest model for predicting axial force. The research indicates that by reconstructing features， the accuracy of axial force predictions can be significantly enhanced compared to direct anomaly detection methods. The R2 values， Mean Absolute Error (MAE)， and Root Mean Squared Error (RMSE) for data with reconstructed features surpass those for data with original features. Particularly， the improvement from an R2 value of 0.921 to 0.956 underscores a significant enhancement in the model's data fitting capability.

Key words: Long Short-Term Memory networks, Random Forest, anomaly monitoring, feature reconstruction

中图分类号: