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

QAREDNN for State Trend Prediction of Rotating Machinery

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。

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.