神经网络理论在谐波测量中的应用

The Application of Neural Network Theory in the Field of Harmonic Measurement

将多层前馈网络用于谐波测量中 ,网络权值的记忆负担主要是来自谐波幅值和相角的变化 ,如果同时改变幅值和相角对网络进行训练 ,由于幅值、相角变化范围大 ,且没有规律 ,研究表明 ,即使使用大量样本 ,训练后的网络将变成杂乱无章不可用 ,因此本文提出了一种方法 ,先对初相角进行确定 ,在初相角已知的状态下 ,再对幅值用基于神经网络理论方法进行测量 ,仿真验证了该方法的有效性。

The memory burden of net weight is from the change of amplitude and phase when the muti layer feedforward neural network is applied to harmonic measurement, If amplitude and phase vary at the same time, the research demonstrates that the whole disciplinal network is unuseful for the wide range of amplitude and phase variance though a lot of sample is used for the network. So this article proposes a new method by which the harmonic phase is determined at first, then the amplitude is measured based on neural network theory. Simulation illustrates the effectiveness of this method.