中国机械工程

• 机械基础工程 • 上一篇    下一篇

用于旋转机械状态趋势预测的量子注意力循环编码解码神经网络

李锋1;程阳洋1,2;陈勇1;汤宝平3   

  1. 1. 四川大学机械工程学院,成都,610065
    2. 航空工业成都飞机工业(集团)有限责任公司,成都,610065
    3. 重庆大学机械传动国家重点实验室,重庆,400044
  • 出版日期:2020-11-10 发布日期:2020-11-16
  • 基金资助:
    国家自然科学基金资助项目(51305283);
    机械传动国家重点实验室开放基金资助项目(SKLMT-KFKT-201718);
    中国博士后科学基金资助项目(2016M602685);
    四川大学泸州市人民政府战略合作项目(2018CDLZ-30)

QAREDNN for State Trend Prediction of Rotating Machinery

LI Feng1;CHENG Yangyang1,2;CHEN Yong1;TANG  Baoping3   

  1. 1. School of Mechanical Engineering, Sichuan University, Chengdu, 610065
    2. AVIC Chengdu Aircraft Industrial (Group) Co., Ltd., Chengdu, 610065
    3. The State Key Laboratory of Mechanical Transmissions, Chongqing University, Chongqing, 400044
  • Online:2020-11-10 Published:2020-11-16

摘要: 提出了基于量子注意力循环编码解码神经网络(QAREDNN)的旋转机械状态趋势预测方法。在QAREDNN中,引入注意力机制以同时重构QAREDNN的编码器和解码器,使QAREDNN能够充分挖掘和重视重要信息,并抑制冗余信息的干扰,从而获得更好的非线性逼近能力;采用量子神经元构建了一种活性值和权值由量子旋转矩阵代替的量子门限循环单元(QGRU),QGRU不仅能够更加精细地遍历解空间,还具有大量的多重吸引子,因此QGRU能代替传统编码器和解码器中的循环单元以提高QAREDNN的泛化能力和响应速度;通过引入Levenberg-Marquardt(LM)法来提高QAREDNN的量子旋转矩阵的旋转角和注意力参数的更新速度。滚动轴承状态趋势预测实例验证了该方法的有效性。

关键词: 量子注意力循环编码解码神经网络, 注意力机制, 量子神经元, 状态趋势预测, 旋转机械

Abstract: A state trend prediction method for rotating machinery was proposed based on QAREDNN herein. In the proposed QAREDNN, the attention mechanism was used to simultaneously reconstruct encoder and decoder of QAREDNN, so that QAREDNN might fully excavate and pay attention to important informations but suppress the interference of redundant informations to obtain better nonlinear approximation capacity. The quantum neuron was used to construct a new quantum gated recurrent unit (QGRU) in which activation values and weights were represented by quantum rotation matrices. The QGRU might traverse the solution space more finely and had a lot of multiple attractors, so it might replace the traditional recurrent units of the encoder and decoder and enhance the generalization ability and response speed of QAREDNN. The Levenberg-Marquardt algorithms were introduced to improve the update speeds of the rotation angles for quantum rotation matrices and the attention parameters of QAREDNN. The effectiveness of the proposed method was verified by the examples of state trend prediction of rolling bearings.

Key words: quantum attention recurrent encoder decoder neural network (QAREDNN), attention mechanism, quantum neuron, state trend prediction, rotating machinery

中图分类号: