多次喷射柴油机实时预测模型参数映射关系辨识

Identification of Correlation Between Parameters in Real-Time Predictive Model for Multi-Injection Diesel Engine

基于经验公式(EF)的参数间映射关系,笔者采用相同的数据和自变量集建立了基于人工神经网络(ANN)的参数间映射关系.分别搭建了EF和基于ANN的柴油机实时物理模型(EF模型和ANN模型),在稳态和瞬态测试循环(WHTC)瞬态工况对比了两种模型的预测性能,结果表明:相比EF模型,ANN模型在稳态工况下对MFB 50、最高燃烧压力(PFP)和平均有效压力(BMEP)的预测精度更好;但是在瞬态工况下,ANN模型预测精度出现明显恶化,低于EF模型.主要原因是模型中参数存在校准误差和不确定性,ANN模型相比EF模型具有更强的非线性拟合能力,但是对参数误差更加敏感.在快速原型设备上测试了两种模型的计算耗时,两种模型计算耗时相当,均约为350μs,满足燃烧过程实时控制的要求.

The empirical function-based(EF-based)correlations between parameters in engine model were obtained in our previous work.In this work,using same data and variables set,artificial neural network-based(ANN-based)correlations were trained and generated.Then,EF-based and ANN-based physics engine model(EF-model and ANN-model)were respectively developed.Predictive performance of the two models under steady-state and world harmonized transient cycle(WHTC)transient operating conditions were verified.The results show that MFB 50,peak firing pressure(PFP)and brake mean effective pressure(BMEP)predictive precision of the ANN-model is better than those of the EF-model under steady operating conditions,but worse under WHTC transient operating conditions.This is mainly because parameter calibration error and parameter uncertainty generally exist in engine model,while the ANN-model is much more sensitive than the EF-model to data’s error and un-certainty even though it is powerful to capture nonlinear characteristics.The computing time of the two models were tested on the rapid prototyping equipment.Test result indicates that computing time of both models are at the same level of 350μs,which meets the computing time requirement for real-time control.