Research On Multi-Objective Optimal Calibration Of Natural Gas Engine Based On Preference MOEA/D Algorithm

Engine multi-objective optimization technology is an effective technical means to reduce engine emissions and improve engine economy,which essentially seeks to optimize the combination of control parameters for multiple indicators such as emissions and fuel consumption.Most of the current studies focus on solving the complete Pareto front in the full range of operating conditions,which is not conducive to the final decision maker’s selection of the preferred solution and wastes computational resources.In addition,the International Maritime Organization(IMO)has set different emission regulations for different regions,so that users have different preferences for engine requirements as the ship sails in different regions.In this paper,a multi-objective performance optimization study with decision maker’s preference information is conducted for 6M33 natural gas engine,aiming to obtain the optimal solution for different decision maker’s preferences and to maintain the best engine performance under different operating conditions.In this paper,a numerical simulation model of a natural gas engine is built using GT-Power software and a bench test is conducted to calibrate and verify the model.On this basis,the effects of ignition timing and air-fuel ratio on engine performance and emissions are studied and analyzed to provide theoretical guidance for test design and subsequent analysis.Based on the limited data,a predictive model for natural gas engines is developed,and the advantages and disadvantages of BP neural network,LSTM neural network and extreme learning machine are investigated and analyzed,among which the extreme learning machine achieves efficient fitting between control parameters and output response due to its powerful nonlinear fitting capability and unique learning training process.Facing the irreconcilable conflict between NOx emission and fuel consumption of natural gas engines,a multi-objective optimization strategy based on the combination of the extreme learning machine prediction model and the MOEA/D algorithm is proposed,and an adaptive evolutionary operator is designed to solve the problem of poor uniformity of the optimal solution distribution.Using this strategy to optimize the natural gas engine,all solutions satisfy the IMO Tier-II emission regulations,and the optimal solution satisfying IMO Tier-III exists.Based on this,a multi-objective optimization of natural gas engine with decision maker preference information is proposed by introducing decision maker preference information into the MOEA/D algorithm framework based on weight vector transformation.Two different preferences are selected for validation.Under the low emission preference scheme,all operating conditions can meet the IMO Tier-III limits,at which the NOx emission decreases by 81% compared with the original engine,while the BSFC increases by 0.35%,indicating that a lower emission level can be obtained with a small sacrifice in economy.When the emission limit is relaxed to IMO Tier-II,the NOx emission is reduced by 35% on average,while the BSFC is reduced by 1.86% compared with the original machine,achieving the optimization of NOx and BSFC at the same time.