Fuel-additives design using hybrid neural networks and evolutionary algorithms

Fuel-additives play an important role in deposit reduction on the valves and combustion chamber of the automobile. They reduce cold-start problems, emissions and improve fuel-efficiency. The testing and design of fuel-additives is an expensive and lengthy procedure. A hybrid first-principles neural-network model for fuel-additive performance prediction was developed in this effort. This model products additive performance directly from the structure of the additive and outperformed existing models based on basic structural descriptors. The design of fuel-additive structures was accomplished using an evolutionary algorithm with problem specific representation and genetic operators. Some results from the model and the design algorithm are discussed in this paper.