Greenhouse gas emission prediction and impact analysis of dual-fuel engine

Abstract

This study explored the profound influence of greenhouse gas (GHG) emissions on global climate change by introducing an innovative prediction model. Utilizing a backpropagation (BP) neural network with dual-hidden layer, optimized through simulated annealing particle swarm optimization (SA-PSO), the model predicted emissions from a diesel/natural gas dual-fuel engine at 1500 rpm across four torque levels: 400, 800, 1200, and 1600 N·m. The inputs included engine torque, injection timing, pressure, excess air coefficient, and natural gas substitution ratio, with CO2 and CH4 as outputs. Evaluation metrics—the coefficients of determination (R²) of 0.9975 for CO2 and 0.9951 for CH4, the root mean square error (RMSE) of 0.062 % and 278.04 ppm, and the mean relative error (MRE) of 0.82 % and 5.35 %, respectively—demonstrated the model’s accuracy. A quantitative analysis using the Mean Influence Value (MIV) algorithm showed engine torque’s pivotal role in emissions at a medium engine speed with contribution rates of 48.5 % and 40.3 % for CO2 and CH4 emissions, respectively. Notably, at a lower load condition (engine torque = 400 N·m), the natural gas substitution ratio was identified as having the most substantial impact on emissions. This study presents a novel approach to predicting and reducing GHG emissions from dual-fuel engines.