Application of the Gaussian Process Regression Method Based on a Combined Kernel Function in Engine Performance...

Area of Science:

• Automotive Engineering
• Computational Modeling
• Machine Learning

Background:

• Current regression methods lack the accuracy needed for advanced automotive simulations like virtual calibration.
• Engine prediction complexity and abruptness pose challenges for traditional modeling approaches.

Purpose of the Study:

• To develop a superior regression modeling method for engine torque, emission, and temperature predictions.
• To enhance the accuracy and applicability of simulation technology in the automotive industry.

Main Methods:

• Proposed a Gaussian process regression model utilizing a combined kernel function.
• Compared the proposed model against linear regression, decision trees, Support Vector Machines (SVM), neural networks, and other Gaussian regression methods.

Main Results:

• The proposed Gaussian process regression model achieved higher prediction accuracy.
• Achieved R-squared values of 1.00 for engine torque and exhaust gas temperature (T4), and 0.9999 for nitrogen oxide (NOx) emissions.
• Demonstrated excellent generalization ability with R-squared values of 0.9993 for torque, 0.995 for temperature, and 0.9962 for NOx on new transient cycle data.

Conclusions:

• The Gaussian process regression method with a combined kernel function offers superior prediction accuracy for automotive engine parameters.
• The model effectively handles the complexity and abruptness of engine predictions.
• Validated high accuracy for performance, temperature, and emission predictions under both steady-state and transient conditions, improving simulation technology.