The fuel consumption rate of Atkinson cycle engine has always been optimized by GT-Power detail model with Genetic Algorithm (GA) (scheme 1),but this method is hard to converge and its calculation is very slow，so a simplified model coupling GA based on Artificial Neural Network (ANN) (scheme 2) for optimization and comparison was proposed.A detailed simulation model based on GT-Power software for Atkinson cycle engine had been carefully built by scheme 1，and the Knock prediction model was built based on Heywood formula.The fuel consumption was then optimized by scheme 1.Scheme 2 used the Latin Hypercube Sample (LHS) to collect 4 500 experimental points，and simplified the GT-Power model and Knock model into the ANN model，which was optimized by the simplified coupled GA model.The results show that by using scheme 2 for the optimization of Atkinson cycle engine，the actual fuel consumption rate is reduced by 4.6%，and the maximum error rate related to the measured optimization results is 7.3%，while the maximum simulation optimization time was saved 322 times that for scheme 1.As a result，the fast global optimization of the fuel consumption rate for the Atkinson cycle engine is feasible by using scheme 2 rather than scheme 1.