【作者】 于静；

【导师】 杨永明；

【作者基本信息】 重庆大学 ， 电气工程， 2003， 硕士

【摘要】 人工神经网络用于波形识别是一个新的研究领域。本文将神经网络这一新方法引入到电网谐波监测中，并就它在电网谐波监测应用中的一系列关键问题进行了较为系统的研究。本论文首先系统的论述了国内外电力网谐波的产生及其危害，以电力系统中主要出现的三次谐波为例，论述了对谐波进行监测的必要性。叙述了国内外的高次谐波测量的主要方法及其优缺点，进而提出应用人工神经网络这一新方法对高次谐波进行监测的优越性。在本文中提出了一种新型电网谐波监测系统模型，通过带通滤波器，采样存储器以及神经网络组成的监测系统可以实时监测电网谐波，当电网中的3次、5次、7次谐波超过容许值时，给出控制信号启动以记录仪记录谐波，并可以控制谐波补偿，最终达到监督电网，抑制谐波、减小谐波的目的。文中详细阐述了监测系统的设计与构想。论述了人工神经网络的模型和学习算法，并围绕本文的研究目的进行了深入的探讨。通过构造一个特殊的多层前向神经网络，利用模拟并行测量装置的基本原理，设计了相应的谐波监测电路。通过对BP算法及其改进型算法的研究，成功的将该算法应用于3次、5次、7次谐波的监测与识别中。论文给出了电路的训练方法和步骤，研究了训练样本的形成方法，并对训练样本的输入方式进行了研究。通过上述的工作，使对解决电网谐波监测与识别问题得到了完善。最后对训练好的神经网络进行了仿真，仿真结果表明本文所提出的基于人工神经网络监测方法的有效性和实用性。本文提出了一种新型电网谐波监测方法，重点研究了用于电网谐波监测的神经网络。虽然只分析了非正弦波形3次、5次、7 次谐波的多层前向神经网络（MLFNN）的训练，以及训练时的样本组成，但该方法对监测其它的谐波也同样适用。如果再有其它类型的畸变波形构成的训练样本来训练本文电路，那么该电路还能够较为精确的监测其它的畸变波形中的谐波。更多还原

【Abstract】 Wave recognition by artificial neural network is a new research field. In order to monitor the power system’s harmonics with new method, the neural network (NN) approach was introduced.It is necessary to monitor the power system’s harmonics because the harmonics do harm to power system. In this paper, we analyze three kinds of typical harmonics in power system. The main method of monitoring of high harmonics is depicted. In addition, the new approach of Artificial Neural Network (ANN) for measuring harmonics on-line is proposed, and the simulation results showed that it is superior to other existing methods. .A new model about measuring harmonics on-line is proposed in this paper. After band pass filtering, sampling storage and the monitoring on-line with artificial neural network (ANN), we can obtain the harmonics signal, which is used to compensate harmonics. Using the basic principle of analog parallel harmonics measurement and a special multi-layer feedforward neural network, a corresponding harmonics measurement on-line is built. <WP=6>In the first application, this thesis discusses the back propagation(BP) algorithm and its modified forms in great detail, and its successful application in monitoring high harmonics (3、5、7). Using the basic principles of analog parallel harmonics measurement and special multi-layer feedforward neural network, the corresponding harmonics measurement on-line is built. Also, the configuration of this Neural Network and the algorithm and the procedure for training the measurement network are given, and the forming of the training samples is indicated. The simulation results prove the approach effective and correct.更多还原