Intelligent regression algorithm study based on performance and NOx emission experimental data of a hydrogen enriched natural gas engine

• QP, NNs and SVM method are respectively used for regression based on calibration experiment data for a HCNG engine.
• BSFC and NOx emissions decrease after hydrogen enriched.
• SVM method shows the best result in prediction of the engine(within 10% for Torque and BSFC, 30% or so for BSNOx in error).

Support vector machine (SVM) method has got rapid development and application because of its advantages in solving problems of small sample regression. In this paper, support vector machine (SVM) method was applied to the engine test data regression analysis. Quadratic polynomial method, neural network and SVM method are respectively used to establish a mathematical model between operating & control parameters and performance parameters based on calibration experiment data for a Hydrogen enriched compressed natural gas (HCNG) engine. Through the comparison of the three methods, SVM method has a higher fitting accuracy than other ways, showing certain superiority in nonlinear system regression. As SVM method is a generic methodology, it may be a new direction for engine calibration algorithm study.