Study of NOx emission for hydrogen enriched compressed natural along with exhaust gas recirculation in spark ignition engine by Zeldovich’ mechanism, support vector machine and regression correlation

Study of NOx emission for hydrogen enriched compressed natural along with exhaust gas recirculation in spark ignition engine by Zeldovich’ mechanism, support vector machine and regression correlation

Support vector machine

Hydrogen

Ammonia

Compressed natural gas

Nitrogen oxides

Regression analysis

Support vector machines

Gases

Ignition

Exhaust gas recirculation

Speed

Sampling

Vectors

NO emission

Hydrogen enriched compressed natural gas

Regression correlation

Thermal NOx emission

Automobile engine manifolds

The hydrogen fuel is an ideal fuel to achieve the zero-carbon emission for heavy-duty vehicles. The experiments have been conducted under a wide range of operating conditions at different ignition timings (0 to 60°CA bTDC), hydrogen fractions (0–40%) in CNG fuel, EGR ratios (0–35.8%), manifold absolute pressures (65–178 kPa), engine speed (700–2000 rpm) at stoichiometric condition. The calibrated quasi-dimensional combustion model has been applied for simulation of in-cylinder pressure and temperature of HCNG engine. These combustion parameters have been utilized in Zeldovich’ mechanism for the prediction of NOx emission. The mean absolute percentage error is 69.1573% and R2 = 0.8353 that is quite high. The support vector machine and regression correlation have been applied for NOx modelling of HCNG engine. The four sets have been formulated at different independent variables: controllable parameters (engine speed, load, ignition timing, EGR ratio and hydrogen fraction in CNG fuel), combustion parameters (maximum pressure, indicated mean effective pressure, coefficient of variation and combustion duration), working medium parameters (inlet temperature, exhaust temperature, manifold absolute pressure and temperature of the coolant) and mixed parameters. The different sets of brake specific NOx emission have been trained by support vector machine algorithm at different independent variables, combinations (5C2, 5C3, 5C4, 5C5), tuning parameters (penalty factor kernel, function width & intensive band loss function) and sample sizes (decrement & increment of sample sizes). The minimum MAPE for SVM is 4.7490% & R2 = 0.9981. The mathematical modelling has been carried out with utilization of suitable independent parameters, combination and sample size. The mean absolute percentage error for regression correlation is 18.2461% and R2 = 0.9619.

123577

318

Fuel

2022

Rao, Anas

Liu, Yongzhen

Ma, Fanhua

journalArticle

0016-2361

10.1016/j.fuel.2022.123577