Multi-objective Optimization Calibration Of Dual-fuel Engine Based On Neural Network And PSO Algorithm

The electronic control system of dual-fuel engine is more complicated,and there are many controlled parameters that need to be calibrated.Therefore,the traditional calibration method cannot achieve the optimization results to meet the multi-targets of emissions and fuel consumption in full working conditions.While the existing engine calibration relies on foreign calibration hardware and software and calibration platform,lacking of independent research.In order to solve the existing problems,this paper adopts the model-based calibration route,introduces the neural network to establish the dual-fuel prediction model,and uses the PSO algorithm to realize the dual-fuel multi-objective optimization calibration.Firstly,the boundary of each input parameter of the engine is determined through the bench test,and the experimental design of dual-fuel is carried out.According to the limitation of dual fuel combustion and emission,the input boundary of each electronic control parameter is determined.Analysis and compare the classical design of experiment,Taguchi experiment design,the space filling experiment design and optimal experiment design four kinds of design of experiment,confirmed with experimental design space filling+V optimization experiment design of mixture experiment design method,the model of variance estimate PEV is about 0.3,the experimental design is reasonable and can reduce the error of the prediction model to some extent.Secondly,the prediction model of dual fuel emission and fuel consumption is established.According to the requirement of modeling,the main parameters of BP and RBF neural networks are set respectively.The differences of the two neural networks in structure,goodness of fit,generalization ability,convergence and training speed were compared.The results show that the RBF prediction model is better than the BP prediction model in many aspects,the goodness of fit value of RBF model is more than 0.98,and the MSE value used to measure the fitting accuracy is 0.001.It is verified that the model has relatively good generalization ability at 10operating points,and the relative prediction error is less than 4%,which can meet the modeling needs of the prediction model.Then,the PSO algorithm is introduced to realize the dual-fuel multi-objective optimization calibration process.The multi-objective optimization problem is described mathematically,the multi-objective problem is transformed into the problem of solving the minimum value of the cost function,and the cost function is taken as the fitness function of the PSO algorithm.In order to solve the problem of premature convergence and poor stability of PSO,the algorithm is improved.In order to reasonably distribute the weight of each optimization parameter,the influence of different weight distribution on the calculated optimization result was studied,and the final weight distribution,namely the weight of NO_x,THC,CO and BSFC,was determined as 0.4,0.2,0.1 and 0.3 respectively based on experience.Finally,the improved PSO algorithm is used to optimize the calculation and generate the control MAP.The operating points with a speed between 800r/min and 1500r/min under the propulsion characteristic curve were verified on the real machine,and the results of the original machine,calculation optimization and actual optimization were compared.The actual results were close to the ideal optimization results,and the actual emissions of NO_x,THC and CO were reduced by 19%,25.6%and 42.9%on average,and the BSFC was reduced by 0.6%on average,achieving the emissions and fuel consumption of multi-objective optimization,promote the characteristic emissions performance indicators to meet IMO TierⅡlimit for NOx in the request.