基于1DCNN-LSTM船舶辅锅炉故障预测方法

Fault prediction of marine auxiliary boiler based on 1DCNN - LSTM

为适应船舶辅锅炉智能化的新要求,提高船舶辅锅炉故障预测的可靠性,采用深度学习方法对船舶辅锅炉进行故障预测研究。在长短时记忆(Long Short-Term Memory,LSTM)神经网络船舶辅锅炉故障预测模型的基础上,采用一维卷积神经网络(1-Dimensional Convolutional Neural Network,1 DCNN)对船舶辅锅炉原始数据进行特征提取和初步分类,以解决LSTM模型在学习数据特征的局限性及对时序数据顺序的依赖性。选取阿法拉伐OS-TCI型船用燃油辅锅炉为研究对象,结合大连海事大学开发的DMS-CSS型轮机模拟器中的船舶辅锅炉仿真模块进行试验研究。采用实船采集的时间序列数据和模拟器采集的故障数据为试验数据来源,分别应用LSTM模型和1DCNNLSTM模型构建船舶辅锅炉故障预测模型并进行了试验对比研究。试验结果表明:1DCNN-LSTM模型相较于LSTM模型其均方误差和均方根误差分别减小了0.0075和0.033,同时减小了拟合误差,提高了预测的可靠性。

Deep Learning Technology is introduced into fault prediction algorithm of marine auxiliary boiler.The long shortterm memory(LSTM)neural network is used to build the prediction model.The source data is processed by 1D Convolutional Neural Network to extract features and do preliminary classification.This process is to overcome the limitation of LSTM model in types of learning data feature and the dependence on the sequence of time series data.Both real data from Alfa Laval OS TCI marine oil fuel auxiliary boiler on board and simulating data set generated by DMS-CSS marine engine simulator are processed by the model for verification.Meantime a LSTM model does the same job for comparison.Experiments show that the mean square error and root-mean-square error of the output from 1DCNN-LSTM model are reduced by 0.0075 and 0.033 respectively compared to those from LSTM model.