Sparse identification with multi-objective optimisation and multi-criteria decision-making for predicting hydrogen–diesel dual-fuel engine performance

Hydrogen enrichment in compression-ignition engines introduces strong nonlinear combustion effects that require advanced modelling for accurate prediction. This study develops a data-driven Sparse Identification of Nonlinear Dynamical Systems (SINDy) framework for predicting hydrogen–diesel dual-fuel (HDDF) combustion across a wide operating range. Systematic evaluation of sparsity levels and 127 function-library combinations showed that hybrid nonlinear libraries provide the most consistent predictive performance. Using all terms achieved R 2 values of 0.9829 (Brake Power), 0.9794 (BSFC), 0.8139 (NO x ), 0.8605 (Torque/BMEP), 0.8641 (BTE), and 0.9938 (UHC), while CO prediction remained challenging due to intrinsic combustion variability. Multi-objective optimisation using Non-dominated Sorting Genetic Algorithm II (NSGA-II) and the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) identified an optimal configuration at λ = 0.05 and θ = 122, balancing BSFC, NO x , and BTE trade-offs. Comparison with artificial neural networks demonstrated improved robustness and generalisation. The framework provides an interpretable and computationally efficient surrogate model for HDDF prediction and optimisation. • A sparse SINDy framework is developed to model hydrogen–diesel dual-fuel combustion. • Model tested across 100 λ values and 127 function-library combinations. • Hybrid nonlinear libraries achieve peak predictive accuracy (R 2 up to 0.9999). • NSGA-II optimisation identifies trade-offs between BSFC, NOx, and BTE. • TOPSIS selects optimal configuration (λ = 0.0797, θ = 91) with balanced R 2 scores.