一种实时电能质量扰动分类方法

发布时间：2018-03-20 21:56

本文选题：强跟踪滤波器　切入点：极限学习机　出处：《电工技术学报》2017年03期 　论文类型：期刊论文

【摘要】：针对电能质量扰动实时分类的需求,提出了一种基于强跟踪滤波器和极限学习机的电能质量扰动分类方法。强跟踪滤波器通过引入渐消因子矩阵克服了扩展卡尔曼滤波器的易发散的问题。强跟踪滤波器不仅可以检测扰动幅值而且还可以提供渐消因子作为特征量,以此识别暂态扰动和谐波。该方法提出使用基波幅值最大值、最小值、波动次数和渐消因子频度均值四个特征量组成特征向量作为极限学习机分类模型的训练样本;最后将分类器用于电能质量扰动识别。为了提高极限学习机分类精度,提出了对少量边界错分样本的类别进行校正的规则校正法。仿真表明改进后的方法能够识别包括两种复合扰动在内的10种电能质量扰动信号,并具有良好抗噪性。与随机梯度下降反向传播方法、最小二乘支持向量机和序贯极限学习机相比,该方法训练和分类速度快,分类准确率高,适合于在线应用。
[Abstract]:According to the demand of real-time classification of power quality disturbance, A power quality disturbance classification method based on strong tracking filter and extreme learning machine is proposed. Strong tracking filter overcomes the divergence problem of extended Kalman filter by introducing fading factor matrix. The waver can not only detect the amplitude of the disturbance but also provide the fading factor as the characteristic. In this method, the maximum and minimum amplitude of the fundamental wave, the number of waves and the frequency mean of the fading factor are used to form the eigenvector as the training samples of the classification model of the extreme learning machine. Finally, the classifier is used to identify the disturbance of power quality. In this paper, a rule correction method for correcting a small number of samples with boundary misalignment is proposed. The simulation results show that the improved method can identify 10 kinds of power quality disturbance signals, including two kinds of complex disturbances. Compared with the stochastic gradient descent backpropagation method, least square support vector machine and sequential extreme learning machine, this method has the advantages of fast training and classification speed, high classification accuracy, and is suitable for online application.
【作者单位】：强电磁工程与新技术国家重点实验室(华中科技大学);长江大学电子与信息学院;湖北工业大学太阳能高效利用湖北省协同创新中心;国网湖南电力公司电力科学研究院;
【基金】：国家自然科学基金(51277080) 太阳能高效利用湖北省协同创新中心科研团队培育项目(HBSZD2014001)资助
【分类号】：TM711

【参考文献】