基于FPGA的神经网络在非线性实时系统辨识中的应用  

Neural Network Application in the Nonlinear Real-Time System Identification Based on FPGA

作　　者：王晓华 王有松 WANG Xiaohua;WANG Yousong(Technology,Siemens Ltd.,China,Beijing 100102,China;Satellite Research and Development Department,Galaxy Space,Beijing 100085,China)

机构地区：[1]西门子(中国)有限公司西门子中国研究院,北京100102 [2]银河航天(北京)通信技术有限公司卫星研发部,北京100085

出　　处：《机床与液压》2022年第20期174-178,共5页Machine Tool & Hydraulics

摘　　要：针对动态系统的辨识问题,设计一种采用现场可编程门阵列(FPGA)实现输出反馈递归模糊神经网络进行非线性动态系统辨识的方法。输出反馈递归模糊神经网络能够实现非线性动态系统的高精度快速辨识,但神经网络复杂的并行结构需要较长的运算时间,限制了它在实时控制和在线系统辨识中的应用。由于FPGA具有并行运算能力,使它在神经网络实现上具有本质的优势。因此,利用NI公司的高性能FPGA板卡以及LabVIEW图形化编程定义FPGA芯片上的逻辑功能,实现具有动态反向传播算法的输出反馈递归模糊神经网络以及舵机位置伺服系统的在线辨识。实验结果表明:基于FPGA的神经网络实现系统辨识周期在毫秒范围内,为非线性实时系统辨识的应用提供了一条新途径。Aiming at dynamic system identification problem,a method of nonlinear dynamic system identification by output feedback recurrent fuzzy neural network was proposed by using a reconfigurable field-programmable gate array(FPGA).The output feedback recurrent fuzzy neural network can realize high-precision and high-speed identification in nonlinear systems,but neural network complex parallelism required a longer time,its application was limited in real-time control and online system identification.According to the parallelism,the FPGA implementation was superior in neural network.Therefore,FPGA logic function was defined by using high-performance FPGA card of NI company and LabVIEW graphical programming.The output feedback recurrent fuzzy neural network with dynamic back-propagation algorithm and the online identification of the rudder position servo system were realized.The results of the experiment confirm that the identification cycle of system can achieve in the millisecond range based on FPGA neural network.It provides a new approach to the application of nonlinear dynamic system identification.

关 键 词：输出反馈递归模糊神经网络 动态反向传播算法 现场可编程门阵列 非线性实时系统辨识 

分 类 号：TP202[自动化与计算机技术—检测技术与自动化装置] TP273.5[自动化与计算机技术—控制科学与工程] TP271.4