Scholarlink

<p>The Drum level is an important parameter that affects the safe and stable operation of the marine supercharged boiler. Predicting the drum level quickly and accurately is beneficial for optimizing control and improving security of the entire system. In this paper, Pearson correlation analysis, maximum mutual information coefficient (MIC) and RF model are used to screen the characteristic parameters of marine supercharged boiler drum level. LSTM and GRU regression models were established, and drum level prediction was conducted based on three feature parameter schemes. The effects of different feature selection methods and neural network models on the accuracy and time consumption of drum water level prediction tasks were explored. The conclusion is that the RF model has the advantage of eliminating redundant parameters when selecting characteristic parameters, so the prediction error can be reduced by more than 50%. Although the LSTM model can maintain a longer information sequence to better cope with the nonlinear lag factors of the marine supercharged boiler parameters, if the feature parameter scheme is appropriate, the GRU model can save 20% of the calculation time while achieving the same accuracy.</p>

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

Comparative study of LSTM and GRU models based on different feature selection methods on the prediction of marine supercharged boiler drum level

  • 期刊论文

  • 发表日期：

    2025年5月22日

查看全文
  • Haotong Wang,
  • Yanjun Li,
  • Jianxin Shi,
  • Yichen Sang,
  • Guolong Li,
  • Sengdi Sun,
  • Xiaojiang Yuan

摘要

The Drum level is an important parameter that affects the safe and stable operation of the marine supercharged boiler. Predicting the drum level quickly and accurately is beneficial for optimizing control and improving security of the entire system. In this paper, Pearson correlation analysis, maximum mutual information coefficient (MIC) and RF model are used to screen the characteristic parameters of marine supercharged boiler drum level. LSTM and GRU regression models were established, and drum level prediction was conducted based on three feature parameter schemes. The effects of different feature selection methods and neural network models on the accuracy and time consumption of drum water level prediction tasks were explored. The conclusion is that the RF model has the advantage of eliminating redundant parameters when selecting characteristic parameters, so the prediction error can be reduced by more than 50%. Although the LSTM model can maintain a longer information sequence to better cope with the nonlinear lag factors of the marine supercharged boiler parameters, if the feature parameter scheme is appropriate, the GRU model can save 20% of the calculation time while achieving the same accuracy.