基于长短期记忆网络的船舶主柴油机性能预测

Performance Prediction of Marine Main Diesel Engine Based on Long Short-Term Memory Network

为预测船舶主柴油机的整体性能,提出一种结合马氏距离和长短期记忆网络(LSTM)的性能预测方法.选取20个常规热力参数作为柴油机的性能参数,引入马氏距离度量不同时刻柴油机的性能退化程度,并归一化为性能指标(PI)序列,以直观描述柴油机的性能退化过程.建立三层LSTM网络模型,分别采用单步法和多步法预测性能指标序列,从而实现柴油机整体性能的趋势预测.以船用柴油机的性能预测实例进行方法验证,性能指标曲线可以直观反映柴油机的性能退化过程,符合柴油机的一般性能退化规律.单步预测的均方根误差(RMSE)和平均绝对误差(MAE)分别等于0.0166和0.0128,多步预测中60步预测的RMSE和MAE分别等于0.0363和0.0315,验证了该方法可用于对柴油机性能的短期波动预测与长期趋势预测.

In order to predict the overall performance trend of marine main diesel engine,a method of long shortterm memory network(LSTM)combined with Mahalanobis distance was proposed.Twenty conventional thermal parameters were selected as the performance parameters of the diesel engine.The Mahalanobis distance was introduced to measure the degradation degree of engine performance at different times,and then normalized to performance index(PI)sequence to describe the degradation process of diesel engine performance visually.A three-layer LSTM network model was established and the PI sequence was predicted by the one-step method and the multi-step method respectively.This method was verified by the performance prediction example of the marine diesel engine.Results show that the PI curve can directly show the performance degradation process of the diesel engine and conform to the general performance degradation law of engine.The RMSE and MAE of one-step prediction are equal to 0.0166 and 0.0128,respectively.In multi-step prediction,the RMSE and MAE of the 60-step prediction are equal to 0.0363 and 0.0315,respectively.It is confirmed that this method can be used for short-term fluctuation prediction and long-term trend prediction of diesel engine performance.