一种基于Volterra频域核的非线性频谱智能表征方法

来源期刊:中南大学学报(自然科学版)2020年第10期

论文作者:陈乐瑞 曹建福 胡河宇 王晓琪

文章页码:2867 - 2876

关键词:BP神经网络;Volterra核;非线性频谱;智能表征

Key words:BP neural network; Volterra kernel; non-linear spectrum; intelligent characterization

摘    要:针对目前基于Volterra核的非线性频谱计算存在的计算量大和准确率低的问题,提出一种基于BP神经网络的非线性频谱智能表征方法。首先,利用递推方法和批量最小二乘方法分别估算出系统的广义频率响应函数(GFRF)幅值和输出频率响应函数(OFRF)幅值;其次,结合非线性频谱特点,将均方根误差( )作为BP神经网络设计指标来确定隐含层神经元数量,利用BP神经网络强大的拟合能力实现各阶频谱幅值的计算;最后,通过机器人驱动系统进行仿真验证。研究结果表明:与常规自适应辨识方法相比,本文方法计算结果与真实结果最接近,且计算速度最高提升了73.30%,进一步证明该方法不但能够满足复杂系统对频谱计算实时性要求,而且可为基于非线性频谱的故障诊断提供精确数据。

Abstract: Aiming at the problem of large amount of calculation and low accuracy of nonlinear spectrum based on Volterra kernel, an intelligent characterization method of nonlinear spectrum based on BP neural network was proposed. Firstly, the generalized frequency response function(GFRF) amplitude and output frequency response function(NOFRF) amplitude of the system were estimated by recursive method and batch least squares method respectively. Secondly, considering the characteristics of the non-linear spectrum, the root mean square error( ) was adopted as the index of BP neural network to design the number of hidden layer neurons. The spectrum calculation of each order was realized by its powerful fitting ability. Finally, the simulation results were validated by the robot driving system. The results show that compared with the traditional adaptive identification method, the results obtained by the proposed method are closer to the true values and the calculation speed is increased by 73.30%, which proves that the proposed method can not only meet the real-time requirement of spectrum calculation for complex systems, but also provide accurate data for fault diagnosis based on non-linear spectrum.

有色金属在线官网  |   会议  |   在线投稿  |   购买纸书  |   科技图书馆

中南大学出版社 技术支持 版权声明   电话:0731-88830515 88830516   传真:0731-88710482   Email:administrator@cnnmol.com

互联网出版许可证:(署)网出证(京)字第342号   京ICP备17050991号-6      京公网安备11010802042557号