CAJ | 학술논문

针对标准BP神经网络收敛速度较慢的问题,本文对所建立的BP网络的学习算法进行了改进,采用LM最优化算法对BP网络进行训练,替代了原来标准BP算法中的梯度下降法寻找最优网络连接权值。燃油喷射系统是柴油机的核心部分,其工作状况直接影响柴油机的燃油过程及其性能,将这种改进的BP算法应用到柴油机燃油故障诊断中,仿真实验证明,该学习算法提高了BP神经网络的学习效率及稳定性,明显地加快了网络的收敛速度,更好地实现了对燃油压力信号的故障诊断。
침대표준BP신경망락수렴속도교만적문제,본문대소건립적BP망락적학습산법진행료개진,채용LM최우화산법대BP망락진행훈련,체대료원래표준BP산법중적제도하강법심조최우망락련접권치。연유분사계통시시유궤적핵심부분,기공작상황직접영향시유궤적연유과정급기성능,장저충개진적BP산법응용도시유궤연유고장진단중,방진실험증명,해학습산법제고료BP신경망락적학습효솔급은정성,명현지가쾌료망락적수렴속도,경호지실현료대연유압력신호적고장진단。
Aiming at the problem of slower convergence rate of the standard BP neural network,in this paper the learning algorithm of established network is improved by LM optimization algorithm which replaces the gradient descent method of the standard BP algorithm to search for the optimal network connection weights.The fuel injection system is the core part of diesel engine,whose working condition directly affect the fuel process and performance of diesel engine.The improved BP algorithm is used in the fault diagnosis of diesel engine fuel.Simulation experiment has proved that this learning algorithm improves the learning efficiency and stability of BP network and obviously improves the convergence speed of the network,and better realizes the fault diagnosis of the fuel pressure signal.