基于神经网络的谐波检测研究

Harmonic Detection Research Based on Neural Network

介绍了一种增加神经网络输入的谐波检测方法,即把负载电流和它在每一周期对应的角度作为BP神经网络的输入,基波电流作为输出,网络的训练精度也相应提高。然后把训练好的网络在Mat-lab/Simulink中仿真,使输出的基波电流与负载电流相减得到谐波电流。该方法与传统的基于瞬时无功功率理论的ip-iq谐波检测方法相比,其在实时性和精确度上有了很大的改善。

A harmonic detection method was presented to take both load current and its corresponding phase in each cycle as the input of BP neural network and the fundamental current as its output, and then to put well- trained network in Matlab/Simulink for simulation so as to get harmonic current by having the fundamental cur- rent and load current subtracted. Compared with traditional i~,-iq harmonic detection method based on instanta- neous reactive power theory proves its greatly-improved property at real time and accuracy. The simulation re- suits indicate that this method can detect harmonic components in the three-phrase circuit at real time and ac- curately, and it has advantages of high precision and little delay.