论文部分内容阅读
为改善单个神经网络模型精度不足的问题,引入有限元建模思想,采用有限元建模和神经网络建模相结合的方法,建立了高压共轨柴油机性能仿真的网络模型,并进行了实验验证。建模过程中,将建模对象空间划分成网格,在节点上利用获取的实验数据分别组成完备和不完备的训练域对网络进行训练,由所有网格上的网络共同完成性能仿真功能,由此建立起一套可用于电控发动机控制系统仿真及参数匹配标定的组合神经网络。实验结果表明:该模型平均输出误差约为2%~4%,使用Pen tium IV 2.0GH z PC机时的计算时间小于60m s,可用于基于模型的发动机控制系统。
In order to improve the accuracy of single neural network model, the finite element modeling method was introduced. The finite element model and neural network model were used to establish a network model of high pressure common rail diesel engine performance simulation. The experimental results . In the process of modeling, the modeling object space is divided into grids, the experimental data obtained on the nodes are used to form the complete and incomplete training domain respectively to train the network, and the network on all the grids is used to complete the performance simulation. Thus a set of combined neural networks that can be used for the simulation of electronic control engine control system and parameter matching calibration are established. The experimental results show that the average output error of the model is about 2% ~ 4%, and the calculation time of the Pentium IV 2.0GH z PC is less than 60m s, which can be used in the model-based engine control system.