基于三角基函数的新型神经网络谐波分析方法

Novel neural network methods for harmonic analysis based on triangle base functions

谐波分析是电能质量检测的关键。改进了基于固定三角基函数的人工神经网络传统模型,仿真验证证实,改进后的模型可以精确获得基波及各整数次谐波的幅值和相位,且直观、收敛速度快;利用Matlab中的自定义神经网络函数创建了一种基于变参数三角基函数的新的人工神经网络模型,配合加窗FFT算法和高效的LM训练算法,能实现准确的整数次和非整数次谐波分析。仿真结果表明,该算法正确,且便于实现,具有一定的实用性。

Harmonic analysis is essential to power quality testing. First, the traditional neural network model based on the fixed triangle base function was modified, which came out to be capable of obtaining precisely the amplitudes and phases of the fundamental and integer harmonic waves by simulations. What＇ s more, it is more visual and has higher convergence speed. Then, a new neural network model based on triangle base function with variable parameters was constructed by using the self-defined neural network function in Matlab, which can be used to carry out accurate anal- ysis for integer harmonics and non-integer harmonics with the help of windowed Fast Fourier transform （FFI＇） algo- rithm and LM training algorithm. Simulation results verified the validity and feasibility of the proposed method.