基于双隐层量子线路循环单元神经网络的状态退化趋势预测

State Degradation Trend Prediction Based on Double Hidden Layer Quantum Circuit Recurrent Unit Neural Network

针对现有人工智能预测方法在旋转机械状态退化趋势预测中存在预测精度较差、计算效率较低等缺点,提出基于双隐层量子线路循环单元神经网络(Double hidden layer quantum circuit recurrent unit neural network,DHL-QCRUNN)的旋转机械状态退化趋势预测方法。首先采用量纲一化排列熵误差构建状态退化特征集,然后将该特征集输入DHL-QCRUNN以完成旋转机械状态退化趋势预测。在所提出的DHL-QCRUNN中,设计双隐层结构以提高网络的非线性映射能力;并引入量子相移门和多位受控非门以实现信息的传递;通过双隐层的量子反馈机制获得输入序列的整体记忆;最后采用输出层激发态的概率幅表示输出,通过以上方法改善了网络的非线性逼近能力和泛化性能,使所提出的旋转机械状态退化趋势预测方法具有较高的预测精度。此外,通过量子Levenberg-Marqudt(LM)算法更新DHL-QCRUNN的网络参数,提高该网络的收敛速度,使所提出的状态退化趋势预测方法具有较高计算效率。滚动轴承状态退化趋势预测实例验证了该方法的有效性。提出了基于DHL-QCRUNN的旋转机械状态退化趋势预测新方法,该方法具有较高的预测精度和较高的计算效率。

In view of the shortcomings such as poor prediction accuracy and lowcomputational efficiency of the existing prediction methodsbased on artificial intelligence in state degradation trend prediction of rotating machinery,a novel state degradation trend prediction method is proposed based on double hidden layer quantum circuit recurrent unit neural network(DHL-QCRUNN).In this method,a normalized permutation entropy error feature set is firstly constructed,and then this feature set is input into DHL-QCRUNN to accomplish state degradation trend prediction of rotating machinery.In the proposed DHL-QCRUNN,a double hidden layer structure is designed to raise the network nonlinear mapping ability;quantum phase-shift gates and multi-qubits controlled NOT gates are introduced to this network for information transfer;the overall memory of input sequences can be obtained by the quantum feedback mechanism in double hidden layers;moreover,the final output is described by the probability amplitudes of excited states in output layer.Therefore,the nonlinear approximation capability and generalization property of network are improved,and then the higher prediction accuracy of the proposed state degradation trend prediction method based on DHL-QCRUNN is obtainedforrotating machinery.Besides,the network parameters can be renewed by the quantum Levenberg-Marquardt(LM)algorithm to improve convergence speed of DHL-QCRUNN,accordingly,higher computational efficiency can be obtained forthe proposed trend prediction method.The example of state degradation trend prediction for rolling bearing demonstrates the effectiveness of the proposed method.A novel state degradation trend prediction method is proposed based on DHL-QCRUNN for rotating machinery,whichownshigher prediction accuracy and higher computational efficiency.