Artificial neural network model development of blended biodiesel

Area of Science:

• Sustainable energy research
• Computational chemistry and engineering
• Environmental science

Background:

• The search for sustainable biofuels is crucial for reducing environmental impact.
• Experimental testing for new biofuels is expensive and resource-intensive.
• Predictive modeling can overcome limitations in experimental biofuel research.

Purpose of the Study:

• To develop an artificial neural network (ANN) model for predicting biodiesel and its blends' emissions.
• To correlate fuel composition, properties, and engine conditions with emission outputs (NOx, COx, HC).
• To provide a cost-effective alternative to physical testing for low-emission fuel discovery.

Main Methods:

• Utilized a comprehensive dataset of 424 biodiesel variations with 17 input variables.
• Optimized an ANN model with a 60/20/20 data division, 0.005 learning rate, batch size 64, 25 hidden neurons, and 200 epochs.
• Employed the Scaled Conjugate Gradient (SCG) algorithm, comparing its performance against the Levenberg-Marquardt (LM) algorithm.

Main Results:

• Achieved a high overall coefficient of determination (R²) of 0.977 and mean squared error (MSE) around 10⁻⁷.
• Demonstrated strong predictive accuracy for feedstock-based predictions (R²=0.980, MSE=2.34×10⁻⁷).
• Showed good predictive capability for blended biodiesels (R²=0.911, MSE=5.29×10⁻⁵), despite data variability.

Conclusions:

• The developed ANN model effectively predicts emissions from biodiesel and its blends.
• This AI approach offers a low-cost, efficient method for discovering novel, low-emission biofuels.
• The model can accelerate the transition towards more sustainable and environmentally friendly fuel options.