基于深层循环神经网络的陀螺仪降噪方法研究

Gyroscope Noise Reduction Method Based on Deep Circulation Neural Networks

陀螺仪固有的随机误差会随时间积累越来越大,循环神经网络作为一种有效处理时间序列信号的算法被广泛使用,然而传统的循环神经网络在处理陀螺仪产生的随机误差上无法解决长期依赖,容易出现梯度消失和梯度爆炸问题.为了获得精确的陀螺仪信号,本文基于循环神经网络变体的长短记忆网络和门循环单元的陀螺仪信号降噪算法,并创新性的将两种网络进行组合验证.文中先是通过Allan方差对陀螺仪随机误差进行误差分析,然后基于LSTM和GRU组合对陀螺仪输出信号进行补偿处理,结果表明LSTM结合GRU对陀螺仪的随机误差处理有明显改善,其中X、Y、Z轴方向陀螺仪的量化噪音、角度随机游走、零偏不稳定性、角速度游走和速度斜坡性能均有不同程度的提升.

The inherent random errors of gyroscopes accumulate more and more over time.Recurrent neural networks are widely used as an effective algorithm for processing time series signals.However,the traditional recurrent neural network(RNN)can not solve the long-term dependence in dealing with the random errors generated by the gyroscope,and it is prone to the problems of gradient disappearance and gradient explosion.In order to obtain accurate gyroscope signals,a de-noising algorithm for gyroscope signals is proposed based on a variant of Long short term memory(LSTM)and gated recurrent unit(GRU).And the two networks are innovatively combined to verify.The random error of the original gyroscope is firstly analyzed through Allan variance,and then the output signal of the gyroscope is compensated based on the combination of LSTM and GRU.The results show that LSTM combined with GRU can significantly improve the random error processing of the gyroscope.X,Y,Z-axis gyroscope’s quantization noise,angle random walk,zero-bias instability,angular velocity walk and speed ramp performance have been improved to varying degrees.