基于量子加权门限重复单元神经网络的性态退化趋势预测

Performance degradation trend prediction method for rotating machinery based on QWGRUNN

提出基于量子加权门限重复单元神经网络(Quantum Weight Gated Recurrent Unit Neural Network,QWGRUNN)的旋转机械性态退化趋势预测方法。采用小波降噪-排列熵法构建性态退化指标集,将该指标集输入QWGRUNN完成旋转机械性态退化趋势预测。QWGRUNN在门限重复单元(Gated Recurrent Unit,GRU)基础上引入量子位来表示网络权值和活性值并构造量子相移门以实现权值量子位和活性值量子位的更新,改善了网络泛化能力,进而提高了所提出的性态退化趋势预测方法的预测精度;采用与自身结构相适应的动态学习参数,改善了网络收敛速度,进而提高了所提出的预测方法的计算效率。滚动轴承性态退化趋势预测实例验证了该方法的有效性。

A novel performance degradation trend prediction method of rotating machinery was proposed based on the quantum weighted gated recurrent unit neural network(QWGRUNN).Firstly,the performance degradation index set for rotating machinery was constructed by using the wavelet denoise-permutation entropy method.Then,this index set was input in to QWGRUNN to accomplish the performance degradation trend prediction of rotating machinery.On the basis of gated recurrent unit(GRU),qubits were introduced in QWGRUNN to represent network weights and activity values,quantum phase-shift gates were constructed to update weight-qubits and activity-qubits,and improve the network generalization capacity and the performance degradation trend prediction accuracy of the proposed method.Finally,the dynamic learning parameter appropriate to the structure of QWGRUNN was adopted to improve the network convergence speed and the computation efficiency of the proposed method.The example of performance degradation trend prediction for rolling bearing verified the effectiveness of the proposed method.