基于CEEMDAN-SVR模型的柴油车氮氧化物瞬态排放预测

喻洋; 王艳艳; 李加强; 刘学渊; 何超; 魏恒; 吉江林

车用发动机 ›› 2021, Vol. 0 ›› Issue (2) : 43-48.

栏目

基于CEEMDAN-SVR模型的柴油车氮氧化物瞬态排放预测

喻洋^1,2，王艳艳^1,2，李加强^1,2，刘学渊^1,2，何超^1,2，魏恒^1,2，吉江林^1,2

作者信息 +

Prediction of NO_x Transient Emission for Diesel Vehicle Based on CEEMDAN-SVR Model

YU Yang^1,2,WANG Yanyan^1,2,LI Jiaqiang^1,2,LIU Xueyuan^1,2,HE Chao^1,2,WEI Heng^1,2,JI Jianglin^1,2

Author information +

文章历史 +

摘要

柴油机具有更高的燃油热效率，但过高的氮氧化物（NO_x）排放限制了其发展，准确预测柴油车在实际道路上工作的排放状态，有助于柴油机的开发和设计。瞬态NO_x排放序列具有非线性、非平稳性和非正态分布的特点，直接使用机器学习算法进行预测变得困难。为准确预测柴油机瞬时NO_x排放，本研究将自适应噪声完备集合经验模态分解（CEEMDAN）应用到柴油NO_x瞬态排放模型中，结合支持向量回归机（SVR）对复杂时间序列的高效建模能力，提出一种柴油机NO_x瞬态预测模型CEEMDAN-SVR。运用CEEMDAN对NO_x排放序列进行分解，获取不同采样频率下的子序列，对各个子序列使用SVR建模预测，然后通过集成各个子序列的预测值获得最终的预测结果。结果表明：CEEMDAN算法能有效降低NO_x序列的非平稳性，通过SVR机器学习模型更容易提取出数据局部特征信息，具有更低的预测误差。

Abstract

Diesel engine has high fuel thermal efficiency, but too high nitrogen oxide(NO_x) emission limits its development. Accurate emission prediction of diesel vehicles in real road conditions is conducive to the development and design of diesel engine. The transient NO_x emission sequence has the characteristics of nonlinear, nonstationary and non-normal distribution, which makes it difficult to predict directly using the machine learning algorithm. In order to accurately predict the transient NO_x emission, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) was applied to the diesel NO_x transient emission model. Combined with the efficient modeling capability of support vector regression (SVR) for complex time series, CEEMDAN-SVR was proposed as a diesel NO_x transient prediction model. CEEMDAN was used to decompose the NO_x emission sequence to obtain subsequences at different sampling frequencies. SVR was used to model and predict each sub-sequence, and then the final prediction result was obtained by integrating the predicted values of each subsequence. The results show that CEEMDAN algorithm can effectively reduce the non-stationarization of NO_x sequence and it becomes easier to extract local feature information of data through SVR machine learning model with lower prediction error.

导出引用

喻洋，王艳艳，李加强，刘学渊，何超，魏恒，吉江林. 基于CEEMDAN-SVR模型的柴油车氮氧化物瞬态排放预测[J]. 车用发动机. 2021, 0(2): 43-48

YU Yang,WANG Yanyan,LI Jiaqiang,LIU Xueyuan,HE Chao,WEI Heng,JI Jianglin. Prediction of NO_x Transient Emission for Diesel Vehicle Based on CEEMDAN-SVR Model[J]. Vehicle Engine. 2021, 0(2): 43-48