本文提出一种新的基于时间序列分解和混合模型的排烟温度预测模型。使用自适应噪声完备集合经验模态分解（Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise，CEEMDAN）将温度序列数据分解为一系列具有不同特征尺度的本征模态函数（Intrinsic Mode Functions，IMFs）和趋势项。通过增项DF单位根（Augmented Dickey-Fuller，ADF）检验判断每个IMF和趋势项的稳定性。采用自回归滑动平均（Autoregressive Moving Average，ARMA）模型对平稳序列进行预测，采用长短期记忆（Long Short Term Memory，LSTM）神经网络模型提取不稳定序列的抽象特征。对各时间序列的预测结果进行重构，得到最终预测值。经实船数据验证，该算法相比于差分自回归滑动平均模型、LSTM模型和CEEMDAN-LSTM模型，RMSE和MAE误差值分别降低约30％和35％，显著提高了主机排烟温度预测精度。
Achieving accurate prediction of the ship's main engine exhaust gas temperature is of great significance for ensuring the safe and efficient operation of the ship, preventing faults, and meeting the increasingly stringent environmental requirements. For this reason, a new exhaust gas temperature prediction model based on time series decomposition and hybrid model is proposed in this paper. The temperature series data is decomposed into a series of Intrinsic Mode Functions (IMFs) and trend terms with different feature scales using Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). The Augmented Dickey Fuller (ADF) method judges the stability of each IMF and trend term. The Autoregressive Moving Average (ARMA) model is used on stationary time series, and a Long Short Term Memory (LSTM) model extracts abstract features of unstable time series. The predicted results of each time sequence are reconstructed to obtain the final predicted value. Verified by real ship data, the algorithm reduces the RMSE and MAE error values by about 30% and 35%, compared with Autoregressive Integrated Moving Average (ARIMA) model, LSTM model and CEEMDAN-LSTM model, and significantly improves the prediction accuracy of exhaust gas temperature of ship main engine.
2025,47(9): 165-169 收稿日期：2024-12-10
DOI：10.3404/j.issn.1672-7649.2025.09.028
分类号：U664.12
基金项目：广西科技重大专项资助项目（AA19254016）
作者简介：江亮（1991-），男，硕士，讲师，高级信息系统项目管理师，研究方向为船舶故障预测与健康管理
参考文献：
[1] ZHANG Y W, ZOU Y J, LIU Z F, et al. Prediction of ship main engine exhaust gas temperature using AR model[J]. Appl. Mech. Mater. 2015, 696: 244-248.
[2] 邹永久. 基于RKGM-AR模型的船舶柴油机热力参数趋势预测研究[D]. 大连：大连海事大学, 2014.
[3] ZHANG Y, LIU P, HE X, et al. "A prediction method for exhaust gas temperature of marine diesel engine based on LSTM[C]//, " 2020 IEEE 2nd International Conference on Civil Aviation Safety and Information Technology, (CASIT) 2020, 49–52.
[4] ZHU L, LI C, WU Q. Ship main engine exhaust temperature prediction based on deep learning algorithm[C]// Proceedings of the 2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC), 632-636.
[5] LI J, LIU Y, GAO Z. Ship exhaust temperature prediction based on attention mechanism and LSTM[C]// Proceedings of the 2020 International Conference on Artificial Intelligence and Big Data (ICAIBD), 78–83.
[6] 焦品博, 王海燕, 孙超, 等. 基于长短期记忆网络的船舶主柴油机性能预测[J]. 内燃机学报, 2021, 39(3): 250-256.
[7] 刘满强, 吴杰. 基于ICEEMDAN与POA-SVM的感应电机故障诊断[J]. 现代制造工程, 2024(5): 127-137.
[8] LI J. "Hybrid model combining EMD and LSTM for marine exhaust temperature prediction " [J]. Ocean Engineering, 2023.
[9] 李洪涛. 基于MI-CEEMDAN-RF-LGBM的风功率预测分析[J]. 电力与能源, 2024, 45(2): 239-242.
[10] 史永胜, 施梦琢, 丁恩松, 等. 基于CEEMDAN–LSTM组合的锂离子电池寿命预测方法[J]. 工程科学学报, 2021, 43(7): 985-994.
[11] 常智慧, 徐耀松. 基于CEEMDAN与INGO优化BiLSTM的短期电力负荷预测 [J/OL]. 控制工程, 1-92024-10-07].
[12] 王瑞, 高强, 逯静. 基于CEEMDAN-LSSVM-ARIMA模型的短期光伏功率预测[J]. 传感器与微系统, 2022, 41(5): 118-122.
[13] 朱利明, 卓静超, 邢世玲. 基于CEEMDAN-NAR-ARIMA组合模型的桥梁结构健康监测应变预测[J]. 科学技术与工程, 2020, 20(4): 1639-1644.
[14] 张雪, 肖秦琨. CEEMDAN组合DISPSO-LSTM的短期电力负荷预测[J]. 西安工业大学学报, 2021, 41(4): 461-469.
[15] 李杰, 孟凡熙, 牛明博, 等. CEEMDAN-WPE-CLSA超短期风电功率预测方法研究[J]. 大连交通大学学报, 2024, 45(2): 101-108.
[16] 张淑娴, 江文韬, 陈玉花, 等. 基于二次模态分解的LSTM短期电力负荷预测[J]. 科学技术与工程, 2024, 24(7): 2759-2766.
[17] 文博, 陈芳芳, 胡道波, 等. 基于完全集成经验模态分解和模糊熵分频的短期风电功率预测[J]. 科学技术与工程, 2023, 23(25): 10835-10845.
[18] 冯建强, 宋昆仑. 基于CEEMDAN-LSTM的桥梁变形时间序列预测研究[J]. 地理空间信息, 2023, 21(7): 40-43.
[19] 简定辉, 李萍, 黄宇航, 等. 基于CEEMDAN-LSTM-CNN网络的短期电力负荷预测[J]. 电工电气, 2023(6): 1-6.
[20] 向玲, 刘佳宁, 苏浩, 等. 基于CEEMDAN二次分解和LSTM的风速多步预测研究[J]. 太阳能学报, 2022, 43(8): 334-339.
