针对在设计电能质量扰动(Power Quality Disturbance,PQD)分类器时人工选取特征过程繁琐并且不够精确的问题,提出一种基于格拉姆角场(Gramian Angular Field,GAF)和卷积神经网络(Convolutional Neural Network,CNN)的PQD分类方法.首先将一维PQD信号映射为二维图像,接着在已有的神经网络基础上构造适用于PQD分类的网络框架.最后将二维图像作为输入,CNN将自动从海量的扰动样本中提取特征并加以分类.仿真结果表明该方法在噪声数据中具有良好的分类性能,是一种行之有效的PQD分类方法.

Improvement of fault location accuracy plays a great role in decreasing repair time and expediting service restoration. However, intricate measurement conditions are produced by some factors including data variation and noise. These factors would lead to nonconformity in reporting samples from multiple detectors and cause errors in the final results calculating. To improve fault location accuracy in transmission grids, an innovative data fusion algorithm, based on the Adaptive Fuzzy Neural Network (AFNN) mechanism, is proposed in this paper. In the model, Fuzzy Data Fusion (FDF) mechanism is formed and serves as the initial fusion to estimate the correction coefficient of Faulted Traveling Wave Propagation Speed (FTWPS). In the final fusion, Adaptive Fuzzy-Neural-Network-based Data Fusion Systems (AFNN-DFSs) are trained to yield the final fault location results with high accuracy. The overall procedure of the proposed fault location technique is constructed using PSCAD/EMTDC and MATLAB. Besides, the feature of the faulted traveling wave is extracted by using Continuous Wavelet Transform (CWT). Finally, case studies and discussions on the new method, based on AFNN-DFSs, are given to prove the advantages of the novel method in computational accuracy.