基于嵌入式的神经网络滤波的研究与实现

Study and implementation of the neural networks filtering based on embedded

在电力系统中谐波检测的实时性和精确度是一个重要部分,目前谐波检测中多以低通滤波器进行滤波,但是由于其要采集大量数据才可以实现稳定输出的固有特性,限制了其实时性和精确度的提高。人工神经网络可以进行滤波分析,但现有神经网络滤波几乎都基于MATLAB仿真,硬件测试的较少。在Linux系统中用C++在Qt软件编程实现三角基函数神经网络滤波,并把文中算法在X86和ARM两个不同芯片下的运行效果与在MATLAB中低通滤波器滤波做对比,发现在实际测试中该算法可以提高精确度,加快响应速度,其次还介绍了算法在基于X86和ARM两个不同芯片下环境配置,编译,实现等技术细节。

The real-time and accuracy are important the low pass filter（LPF） is usually method, but due part of the harmonic detection in the power system, to the inherent characteristics that it has to collect large amounts of data in order to achieve stable output, limits the improving of the real-time and accuracy. Artificial neural network can detect harmonic analysis, but the existing neural network harmonic are almost based on the MATLAB simulation, the hardware test is less, this article in Linux system with C ＋ ＋ in Qt programming realization of harmonic detection of triangle basis function neural network, and the running effect of this algorithm in X86 and ARM two different chips compares with the running effect of low pass filter in harmonic detection to in MATLAB, found in the actual test, the algorithm can improve accuracy and response speed. It also introduces the algorithm in the X86 and ARM two different chip under the environment configuration, compiler and the technical details.