论文部分内容阅读
提出采用投影寻踪网络逼近Volterra滤波器的方法以实现光纤陀螺输出信号非线性噪声消除.投影寻踪网络采用批量学习和参数交替优化的训练算法,可以自适应确定神经网络的规模、参数和神经元函数,具有简捷的网络结构和较强的鲁棒性,克服了Volterra滤波器随着阶数增加,滤波系数的数量呈几何级数增长,实现困难的问题.仿真结果表明PPLN网络滤波器比Volterra滤波器具有更好的消噪效果.
The method of Projection Pursuit Network approximation to Volterra filter is proposed to realize the nonlinear noise cancellation of the output signal of fiber gyroscope.The Projection Pursuit Network adopts the training algorithm of batch learning and parameter alternation optimization to adaptively determine the size of neural network, Element function has a simple network structure and strong robustness, overcomes the Volterra filter as the order increases, the number of filter coefficients increased geometrically, difficult to achieve.The simulation results show that PPLN network filter ratio Volterra filters have better noise reduction.