多尺度稀疏电能质量扰动识别方法

针对传统电能质量扰动识别中存在数据量大、扰动特征依赖主观选择的问题，提出一种多尺度稀疏电能质量扰动深度识别方法. 首先，构建电能质量的多尺度稀疏模型，通过对扰动信号平稳小波多尺度变换获得扰动的低高频信息；然后，对其压缩采样获得降维的测量数据，并在此基础之上，应用正交匹配追踪算法求取各层稀疏系数组成稀疏向量，将稀疏向量输入深度置信网络，实现扰动的智能识别；同时，为进一步提高网络识别的准确性，采用交叉熵算法完成对网络隐含层数、学习率等参数寻优；最后，为验证所述方法的有效性，针对几类典型的单一扰动和复合扰动信号进行大量仿真试验. 结果表明:在理想环境和噪声环境下，针对七类典型单一扰动，平均识别率达到99.0%和96.71%以上；针对13类多重扰动，平均识别到达97.69%和94.62%以上. 

Recognition Method for Multi-scale Sparse Power Quality Disturbance

In the traditional power quality disturbance recognition, there is a large amount of data and disturbance characteristics are dependent on subjective selection. To deal with these problems, a recognition method for multi-scale sparse power quality disturbance is proposed. Firstly, a multi-scale sparse model for power quality signal is constructed. Through the stationary wavelet transform (SWT) for the disturbance signal, its low and high frequency information is obtained. Then by compressed sampling for the disturbance signal, the dimension reduction data are obtained. Further, sparse coefficients calculated by orthogonal matching pursuit (OMP) algorithm constitute a sparse vector, which is directly inputted into the deep belief network to achieve intelligent disturbance classification. Meanwhile, to improve the recognition rate, cross-entropy algorithm is applied to find the optimal parameters such as the number of hidden layers and learning rate. Finally, in order to verify the effectiveness of the proposed method, a large number of simulation tests were performed for several typical single disturbances and mixed disturbances. The simulation results demonstrate that in the ideal environment the averaged recognition rate of this method for seven typical single disturbances and thirteen mixed disturbances is 99.0% and 97.69% respectively, and in noisy environment at least 96.71% and 94.62% respectively, which shows that the proposed method has a desirable performance in disturbance identification.