Study On Model-based Emissions Prediction Of Gasoline Engines And Real Drive Emission Characteristics

With the increasingly stringent ChinaⅥemission regulations,the engine development stage is faced with bench emission calibration,vehicle WLTC cycle and other laboratory tests and RDE road emission tests,also accompanied by the problems of long test cycle and low development efficiency.Model-based engine emission prediction method can predict unknown engine emissions in the case of known partial data,which greatly reduces the workload of laboratory testing and becomes one of the technologies attracting much attention at present.However,due to the influence of many non-vehicle factors on emissions in RDE road test,it is impossible to reduce the test workload by model-based prediction method.Therefore,the current research focus is to study the influence of non-vehicle factors on RDE emissions to guide the development of engine emission calibration.In order to reduce engine development workload and improve development efficiency under the background of ChinaⅥemission regulations,based on type I and type II tests in ChinaⅥemission regulations,this paper constructed gaussian process regression optimized by particle swarm optimization algorithm(PSO-GPR)model for gasoline engines and light-duty vehicles,and the emission prediction from engine steady state to vehicle WLTC transient was studied based on the model.At the same time,the influence of non-vehicle factors on RDE emissions was studied,and the reasons for the emission differences were analyzed from the engine level,so as to explore the possibility of improving engine development efficiency under the background of ChinaⅥemission regulations.First,based on engine ignition angle calibration test data,with engine speed,torque and ignition angle as input parameters,the interval filling method based on PSO-GPR model was used to predict HC,NOx and CO of remaining calibration data using about 50%calibration data as training set,and compared with traditional GPR model and multivariate polynomial regression(MPR)model.The results show that the R~2 of PSO-GPR model for predicting the three kinds of emissions are all greater than 0.97and prediction results are better than GPR and MPR models;The prediction accuracy of the three kinds of emissions increases with the increase of the amount of data in the training set,and the sensitivity to the amount of data from high to low are CO,HC and NOx;The prediction R~2 of PSO-GPR model for the other three gasoline engines with different displacement are all greater than 0.95,indicating that the model is universal.Then,transient emission prediction of PSO-GPR model was verified based on vehicle hub emission test,the self-designed training set was used to predict the original emissions of WLTC cold start stage and the design method of training set was studied based on the model.The generalization ability of the model was further verified by extending the prediction to the WLTC transient NOx original emission.The results show that the model have a good prediction effect on NOx and CO in the cold start stage of WLTC,and the prediction R~2 is 0.9089 and 0.8103.The prediction R~2 of WLTC transient NOx emission is about 0.9,higher than that of BP neural network(0.8102).It was found that when the range and proportion of engine parameters in the training set were closer to the prediction set,the prediction effect was better.Finally,based on RDE road emission test,the influence of driving state,ambient temperature and altitude on RDE emission was studied.The results show that RDE emissions increase with the aggressive driving state,especially CO and PN emissions.The emissions in the aggressive driving state are 8.78 and 11.55 times of that in the normal driving state;The CO emission of RDE increases with the decrease of the ambient temperature,and the NOx emission decreases with the decrease of the ambient temperature.The CO and NOx emission at-7℃are 2.34 and 0.54 times of that at 10℃,while PN does not change significantly with the ambient temperature;RDE emissions at 2400m are higher than those in plain areas and CO,NOx and PN are 1.61,1.15 and2.01 times of the plain area.