用于钢丝绳断丝定量识别的神经网络模型参数优化

Parameter Optimization of Neural Network for Quantitative Identification of Broken Wire Rope

建立了钢丝绳断丝定量识别的BP神经网络模型,重点从网络输入特征值的分析与优化、网络训练集与测试集的合理选择、网络训练目标的确定3个方面讨论了优化神经网络参数与性能的方法。经实际网络的训练及测试,证明了合理参数的选择改进了网络性能,提高了钢丝绳断丝定量识别的精度,具有实际工程意义。

For quantitative identification of broken wire rope,a BP neural network model is established.This paper is focused on eigenvalue analysis and optimization of neural network,selection method of network training set and test set,the selection of network training objective,and from this three aspects the method to optimize network parameters and performance is discussed.Through actual trainings and tests of neural network,it is demonstrated that optimization of network parameters can improve performance of neural network,enhance ability for quantitative identification of broken wire rope,and has practical engineering significance.