Predictive study of ultra-low emissions from dual-fuel engine using artificial neural networks combined with genetic algorithm

Abstract

ABSTRACTMany degrees of freedom on engine operating parameters limit the optimizing of engine managements for the sake of simultaneously complying with emission regulations and energy economy requirements. Adaptive neuro-fuzzy inference system (ANFIS) is the combination of neural network and fuzzy logic, able to solve nonlinear problems those do not have algorithmic solutions and cannot be modeled mathematically, thus eliminating the limitations of classical approaches. In this study, ANFIS was employed to map the relationships between controlled boundaries and engine performances. A total number of 80 experimental data on dual-fuel diesel engine were selected for training and testing the ANFIS model which has six input variables (diesel fuel injection timing, gasoline premixed ratio, rate of exhaust gas recirculation, indicated mean effective pressure, and the timings of 10% and 50% of total heat release) within a wide validity ranges of engine operating parameters and four outputs of engine emissions and performance. Then, the ANFIS outputs were used to evaluate the objective functions of the optimization process, which was performed with a genetic algorithms (GA) multi-objective optimizing approach. Finally, the Pareto-optimal sets were plotted with minimized NOx as well as soot emissions within the imposed constraints of pressure rise rate and efficiency. This paper studied the feasibility of using ANFIS in combination with GA to optimize the diesel engine settings so that the optimal engine performance and emission behavior would be obtained. The characteristics of the optimal solutions were ultimately explored by sensitivity analysis.