Predicting performance and emissions of dual-fuel engine under pilot-injection strategy using a hybrid BO-IPSO-BP neural network model

The automotive industry faces increasing pressure to reduce greenhouse gas emissions and enhance energy efficiency, driven by global carbon neutrality objectives. Dual-fuel (diesel/natural gas) engines have emerged as a promising solution due to their potential for reducing emissions and improving fuel economy. However, accurately predicting their performance and emission profiles remains a challenge. Current artificial neural network (ANN) models are hindered by inconsistent input selection, the exclusion of critical injection parameters (e.g., pilot-injection timing), and suboptimal hyperparameter tuning, all of which undermine their accuracy and reliability. These limitations obstruct effective engine calibration and emission reduction efforts, which are essential for promoting environmental sustainability. This study introduces a hybrid BO-IPSO-BP neural network model to address these issues. By integrating Bayesian Optimization (BO) for hyperparameter tuning, Improved Particle Swarm Optimization (IPSO) for weight initialization, and Backpropagation (BP) for precise training, the model enhances predictive capabilities. Trained on 218 experimental data points from a 6-cylinder turbocharged engine, the model incorporates key inputs such as injection timing and air-fuel ratio. The results demonstrate its superior performance: test R 2 values exceeded 0.976 for performance parameters (BTE, BSFC) and 0.988 for emissions (CO, NO x , HC). Notably, the testing RMSE for BTE and BSFC was reduced by 28.1 % and 24.1 %, respectively, compared to the BO-BP baseline, and the MAE for CO emissions was reduced by 27.3 %. Overall, 88.6 % of predictions had an error of less than 10 %, with no high-risk errors (>25 %). This model offers a robust and accurate tool for predicting dual-fuel engine behavior, addressing the need for reliable performance and emission forecasts. It contributes to advancing engine calibration, emission mitigation, and sustainable machinery design, while supporting global decarbonization efforts.