基于神经网络的费托燃油发动机特性研究

Research on FT fuel engine characteristics using neural network

基于反向传播(Back Propagation,BP)神经网络强大的非线性逼近和自学习能力,设计3层网络模型,采集发动机台架试验数据作为样本进行模型训练和检验.以发动机转速、转矩、供油提前角和以天然气为原料的费托燃油(GTL)与柴油混合燃料特性参数十六烷值、硫含量、芳香烃含量为输入,建立BP神经网络模型预测GTL发动机特性.结果表明,采用该模型可同时预测GTL发动机功率、油耗、排温、HC、CO、CO_(2)、NO_(x)和碳烟排放等特性;与试验数据对比,预测结果的相对误差基本在5%以内,表明该模型具有较高的模型精度和良好的泛化能力.

Based on the strong nonlinear approximation and self-learning ability of Back Propagation(BP)neural network,a three-layer network model was designed,and the engine bench test data were collected as the samples to train and verify the model.The engine speed,torque,fuel supply advance angle and the characteristic parameters of the fuel mixture of Fischer-Tropsch gas-to-liquid(GTL)and diesel,such as cetane number,sulfur content and aromatic hydrocarbon content,were taken as inputs.The BP neural network model was established to predict the characteristics of GTL engine.The results show that the model can predict the power,fuel consumption,exhaust temperature,HC,CO,CO_(2),NO_(x) and soot emission of GTL engine at the same time.Compared with the experimental data,the relative error of the prediction results is nearly within 5%,which shows that the model has high model accuracy and good generalization ability.