Research On Engine Calibration Model Using Gaussian Process Regression

With the booming development of the global automotive industry,petroleum resource is depleting,and environmental pollution is becoming serious increasingly.Many countries have introduced stricter emission regulations and laws successively,which have put forward higher requirements on the economic and emission performance of automobiles.In order to cope with such problems,the automotive engine electronic control technology continues to develop and progress,the electronic control system is more advanced and complex,the parameters that the engine needs to be optimized for calibration are increasing constantly,and the coupling relationship between different parameters is becoming more and more complicated.The traditional optimization calibration technology can no longer meet the calibration needs of modern engines.The model-based engine calibration technology is generated and developed into an important technical methods in the field of engine calibration.Under the background,this paper takes a highpressure common rail diesel engine as the research object,and aims to achieve the optimal economic efficiency when the emissions meet the requirements of regulations.The model-based engine calibration technology is studied and developed,and the main research contents are as follows:An engine calibration model based on Gaussian Process Regression(GPR)is constructed.On the basis of formulating the Design of Experiment(DoE)scheme and completing the engine bench tests,the GPR model is used to model the test data.By comparing the performance of different covariance functions,the square exponential covariance function is selected as the best.The Newton gradient method is used to obtain the optimal hyper-parameter.The result of performance evaluation shows that the GPR model constructed here is accurate and reliable,and can be used for calibration and optimization of engine.The prediction results of the GPR model and the commonly used models in engine calibration are compared.The quadratic polynomial,cubic polynomial and neural network models are selected and constructed,and compared with the GPR model.The result shows that the GPR model has higher prediction accuracy and provides confidence interval of parameters,compared with the commonly used calibration model.Using the constructed GPR model,a virtual calibration is performed to obtain the prediction data for each operating point.Then the engine control parameters are optimized locally to obtain the initial MAP,and the linear interpolation method is used to obtain smoother and closer MAP.The MAP is re-introduced into the ECU for engine bench tests,and the result shows that the optimized fuel economy area is expanded significantly and the economic performance of engine is improved effectively.