IEEE Trans Neural Netw Learn Syst. 2018 May;29(5):1598-1607. doi: 10.1109/TNNLS.2017.2677961. Epub 2017 Mar 16.
This paper proposes an application of a least mean-square (LMS)-based neural network (NN) structure for the power quality improvement of a three-phase power distribution network under abnormal conditions. It uses a single-layer neuron structure for the control in a distribution static compensator (DSTATCOM) to attenuate the harmonics such as noise, bias, notches, dc offset, and distortion, injected in the grid current due to connection of several nonlinear loads. This admittance LMS-based NN structure has a simple architecture which reduces the computational complexity and burden which makes it easy to implement. A DSTATCOM is a custom power device which performs various functionalities such as harmonics attenuation, reactive power compensation, load balancing, zero voltage regulation, and power factor correction. Other main contribution of this paper involves operation of the system under abnormal conditions of distribution network which means noise and distortion in voltage and imbalance in three-phase voltages at the point of interconnection. For substantiating and demonstrating the performance of proposed control approach, simulations are carried on MATLAB/Simulink software and corresponding experimental tests are conducted on a developed prototype in the laboratory.
本文提出了一种基于最小均方（LMS）的神经网络（NN）结构在三相配电网异常情况下的电能质量改善中的应用。它在配电静止补偿器（DSTATCOM）中使用单层神经元结构进行控制，以衰减由于连接多个非线性负载而注入电网电流中的噪声、偏差、陷波、直流偏置和失真等谐波。这种基于导纳 LMS 的 NN 结构具有简单的架构，降低了计算复杂度和负担，使其易于实现。DSTATCOM 是一种定制功率装置，可执行各种功能，如谐波衰减、无功功率补偿、负载平衡、零电压调节和功率因数校正。本文的其他主要贡献还包括在配电网异常情况下的系统运行，即在互联点处电压中的噪声和失真以及三相电压不平衡的情况下的运行。为了证实和证明所提出的控制方法的性能，在 MATLAB/Simulink 软件上进行了仿真，并在实验室中对开发的原型进行了相应的实验测试。
