| 周圣凯,寇传富,叶宇,等.基于NARX神经网络的瞬态虚拟排温传感器[J].内燃机工程,2025,46(5):69-75. |
| 基于NARX神经网络的瞬态虚拟排温传感器 |
| A Transient Virtual Exhaust Temperature Sensor Based on NARX Neural Network |
| DOI：10.13949/j.cnki.nrjgc.2025.05.008 |
| 关键词:外部输入非线性自回归模型 神经网络 瞬态 柴油机 排气温度 虚拟传感器 |
| Key Words:nonlinear autoregressive model with exogenous input（NARX） neural network transient diesel engine exhaust temperature virtual sensor |
| 基金项目:广西科技重大专项项目（AA23062041） |
|
| 摘要点击次数: 394 |
| 全文下载次数: 231 |
| 摘要:基于台架采集数据，采用外部输入非线性自回归（nonlinear autoregressive model with exogenous input， NARX）神经网络建立了具备瞬态特性的柴油机排气温度计算模型作为虚拟传感器，并采用并发式训练方法对模型进行训练。将结果与前馈神经网络、长短期记忆网络（long short term memory， LSTM）神经网络及量产发动机的排温传感器采集结果进行对比。经验证，稳态工况下，两种神经网络均能达到较高精度；欧洲瞬态循环 （European transient cycle， ETC）工况下，NARX神经网络计算温度的最大偏差为6.6 ℃，量产发动机排温传感器测得温度最大偏差为45.9 ℃。NARX神经网络所需的计算时间约为现有电控单元排温模型的2.5倍。 |
| Abstract:Based on the data collected by test bench， a diesel engine exhaust temperature computational model with transient characteristics was established by using nonlinear autoregressive model with exogenous input （NARX） neural network as a virtual sensor， and the model was trained by batch training method. The results were compared with the feedforward neural network， long short term memory （LSTM） neural network and the collected results of the exhaust temperature sensor of production engine. The results show that both neural networks can achieve high precision under steady states. Under the European transient cycle（ETC）， the maximum deviation of the NARX neural network computation temperature was 6.6 ℃， and the maximum temperature deviation measured by the production engine exhaust temperature sensor was 45.9 ℃. The computation time required for the NARX neural network was approximately 2.5 times that of the exhaust temperature model in the existing electronic control unit（ECU）. |
|
查看全文 HTML
查看/发表评论 下载PDF阅读器 |