Artificial neural network models for forecasting the combustion and emission characteristics of ethanol/gasoline DFSI engines with combined injection strategy

Abstract

Ethanol was a viable alternative and renewable fuel for spark ignition engines, and ethanol/gasoline dual-fuel combustion with combined injection strategy can strengthen the benefits of using ethanol in engines. However, delicate calibration and optimization were required for ethanol/gasoline DFSI engines due to their higher control complexity and flexibility. Therefore, Artificial Neural Network (ANN) models were developed to represent the performance of DFSI engines. This study established the architecture for the ANN models of ethanol/gasoline DFSI engines, and the topology of 8-36-24-3, 8-64-70-6, 8-48-48-3, and 8-36-16-2 were suggested for the modeling of performance, combustion characteristics, gaseous and PN emissions, respectively. The non-linear relationship between the core control variables and performance, combustion, and emission characteristics of ethanol/gasoline engines can be accurately mapped by the proposed ANN models. The regression values were within the range of 0.9387–0.9962, and the mean square relative errors were within the range of 0.000184–0.03935 between the ANN predicted and experimentally measured results. Moreover, the ANN models had the advantages of high accuracy, sound model completeness, superior robustness, and satisfied reliability, which were desirable for the calibration and optimization of engines.