摘要
传统BP神经网络是解决电容层析成像系统流型辨识经典的算法,虽然在一些简单问题上达到了工业实际应用的要求,但如果解决复杂工业问题时就会暴露出很多缺陷。针对传统BP神经网络算法的不足,为降低误差震荡现象,引入了自适应调节学习速率和附加动量因子。通过输入电容值进行训练,得到适合流型识别神经网络。仿真实验结果表明,该算法不仅继承传统BP神经网络的优点,而且还提高了ECT系统流型辨识中的收敛速度慢,解决了容易陷入局部极小值的问题。
Traditional BP neural network is a typical mehtod to solve ECT system of flow pattern identification.It is applied to the simple problems in industrial applications,but there are many defects in solving complex industrial problems.In this paper based on the analysis of deficiency of BP neural network,for reducing the error oscillation,the adaptive learning rate adjustment factor and the additional momentum is introduced.In this method,the electrical capacitance values are input to train a network to identify the flow patterns.The simulation results show the algorithm not only inherits the advantages of traditional BP neural network,but also improve slow convergence and solve being prone to fall into local minimum problems in flow pattern identification of ECT system.
出处
《哈尔滨理工大学学报》
CAS
北大核心
2018年第1期105-110,共6页
Journal of Harbin University of Science and Technology
基金
国家自然科学基金(60572153
60972127)
黑龙江省博士后资助项目(LBH-Z11109)
黑龙江省青年科学基金(QC2012C059)
黑龙江省教育厅科学技术研究项目(11541040
12511097)
关键词
电容层析成像
流型辨识
BP神经网络
局部极小值
收敛速度
electrical capacitance tomography
flow regime identification
BP neural network
local minimum
convergence speed
