Development of hybrid DNN-iPOA model for predicting diesel engine performance and emission for biodiesel optimization

This paper proposes a new method as a hybrid deep learning-based framework, the Deep Neural Network- Improved Pelican Optimization Algorithm (DNN-iPOA) framework, to improve the prediction of the engine performance and emission of biodiesel-fueled diesel engines. The model uses a 201 experimental data set having five important input variables (density, kinematic viscosity, cetane number, flash point and net calorific value) to predict six important output parameters: Brake Thermal Efficiency (BTE), Brake Specific Fuel Consumption (BSFC), Carbon Monoxide (CO), Nitrogen Oxides (NOX), Unburned Hydrocarbons (HC) and smoke opacity. Current performance of the proposed DNN-iPOA model is remarkable, with Mean Absolute Error (MAE) = 0.0064, Mean Squared Error (MSE) = 0.0001, Root Mean Squared Error (RMSE) = 0.008, Normalized MSE (NMSE) = 0.0035, Coefficient of Determination (R2) = 0.9965, and Adjusted R2 = 0.9964. The proposed DNN-iPOA model showed significant performance, reducing the prediction error of MAE by about 2.8 % in smoke emissions. It also increased the accuracy of R2 by up to 65.4 % in key parameters such as smoke emissions when compared to conventional models such as Recurrent Neural Network (RNN), Feedforward Neural Network (FNN), Convolutional Neural Network (CNN), and Long Short Term Memory (LSTM). These findings validate the better accuracy, generalization, and robustness of the model; hence, a valid model in the maximization of biodiesel utilization and minimization of diesel engine emissions.