神经网络方法在发动机特性仿真中的应用研究

Investigation on Neural Network Methods Applied in Engine Performance Simulation

根据发动机特性实验结果研究了发动机扭矩和转速与燃油耗及节气门开启角之间的关系.分析了利用BP神经网络工具箱对该网络进行设计的过程,建立了发动机扭矩和转速与燃油耗及节气门开启角关系的神经网络模型,即基于函数trainbpx和trainlm的模型.结果表明,基于函数trainbpx的模型的稳定性较好,而基于函数trainlm的模型的结构较简单,训练时间较短.

The relationships among the torque, the speed, the specific fuel rates of the engine, and the throttle angle were studied on the basis of the engine performance test results. The back-propagation network （BP） and its design with the toolbox in the MATLAB were analyzed. Using the functions of trainbptx and trainlm, we established the neural network models for the above mentioned relationships. The results indicate that, the stability of the model with trainbpx is better than that of the model with trainlm, while the model with trainlm has the simpler configuration and shorter training time than those of the model with trainbpx.