王立宇，滕勤*，庄远
合肥工业大学汽车与交通工程学院，安徽合肥230009
摘要：基于反向传播（back propagation，BP）神经网络的逼近能力和自学习能力，以汽油机工况参数、状态参数和喷水控制参数为输入，分别建立多输入/单输出的点火提前角、油耗和排放预测模型；以试验设计和发动机台架的实测数据为基础构建、训练模型，用均方差评价网络训练与拟合效果；用相关系数判断输出值与目标值之间的密切程度；用决定系数和相对误差检验模型精度。结果表明:在训练、验证、测试、整体检验阶段，3种预测模型线性相关系数均大于0.8，均方差均小于0.01，决定系数都逼近于1，相对误差均小于15%，建立的模型能够较好预测喷水汽油机不同工况的性能。
关键词：喷水汽油机；性能预测；BP神经网络模型；拟合
Abstract:Based on the approximation and selflearning capabilities of the back propagation (BP) neural network, a multiinput/singleoutput ignition angle prediction model, a fuel consumption prediction model, and an emission prediction model are developed with gasoline engine operating parameters, condition parameters, and water injection control parameters as inputs, respectively.These models are constructed and trained by collecting the measured data of the engine frame based on the experimental design, and the training and fitting effects of the network are evaluated by correlation coefficients and mean squared errors. The influence of input and output variables are evaluated by coefficients of determination, and the accuracy of the models are checked by relative errors. The results show that the linear correlation coefficient of the three prediction models is greater than 0.8, the mean squared deviation is less than 0.01, the coefficient of determination is close to 1, and the relative error is less than 15% in the phases of training, validation, testing, and overall inspection, and the model can predict the performance of gasoline engine under different operating conditions.
Keywords:waterinjected gasoline engine；performance prediction；BP neural network model；fitting
|
