Predictions On Laminar Burning Velocity Of Dimethyl Ether/Air Mixtures Using Machine Learning

As one of the basic combustion characteristics of a fuel,laminar burning velocity is the basis for turbulent combustion research and the important basic input parameter for engine simulation.Accurate measurement and prediction of the laminar burning velocity of fuel is of great significance to the design and improvement in engines.Based on experimental and simulation data,laminar burning velocities of DME were modeled,predicted and analyzed using machine learning multivariable algorithm,neural network algorithm and genetic optimization algorithm.The relationship between laminar burning velocity of DME and initial conditions was founded,and three prediction models and prediction results were obtained.The prediction models of DME laminar burning velocity established in this paper can provide simple and accurate input data for the numerical simulation od DME engine,thus saving the research cost and calculation time.The main research work of the thesis is reflected in the following aspects:1)Based on the experimental data in the literature,the detailed chemical reaction kinetic mechanism of dimethyl ether and the numerical simulation software Chemkin Ⅱ,168 sets of dimethyl ether/air premixed laminar burning velocity under different initial conditions were obtained,providing sample data for the establishment of machine learning multivariate regression model,neural network model and genetic optimization algorithm model.2)Based on the multivariate regression algorithm of machine learning,a functional relationship between premixed laminar burning velocity of dimethyl ether/air mixture and initial conditions(initial temperature,initial pressure,and equivalent ratio)was established.It is found that the predicted fitting degree of the regression model is 0.973.Laminar burning velocity of DME shows a negative exponential relationship with the initial pressure while a positive exponential relationship with the initial temperature,indicating that the laminar burning velocity of dimethyl ether decreases with the increase of initial pressure while increases with the increase of initial temperature.3)A three-layer neural network model for the prediction of laminar burning velocity of DME is established using neural network algorithm of machine learning.Effects of number of neurons in hidden layer on the prediction ability of the neural network prediction model were analyzed.It is found that the model prediction gets the best ability as the number of neurons in hidden layer is 5,with the fitting degree of 0.991.4)Based on the intelligent optimization genetic algorithm,the initial weight threshold of the neural network prediction model was optimized,and a more stable and accurate genetic algorithm optimization model was founded,with the fitting degree of 0.998.By comparing the prediction abilities of the three models,it is found that the neural network model and the genetic algorithm optimization model have better prediction performance for the sample data,while the multivariate regression model has better prediction ability under normal or high pressure.