In order to meet the requirements of classifying power-quality disturbances in real time,this paper proposes a new method based on the strong tracking filter (STF) and the extreme learning machine (ELM).STF is the modified version of the extended Kalman filter (EKF) by introducing the fading factor matrix to solve the problem of divergence.STF can not only detect the amplitude of the fundamental but also provide the fading factor as a feature identifying transient disturbances and harmonics.The proposed feature vector sets were composed of four features including the maximum and the minimum of the fundamental amplitude,the number of fluctuations,and the mean value of the fading factor frequentness.They were input into the ELM as the training examples to obtain a classifier for identifying disturbances.In addition,some rules were used to correct the error classification in a few boundary samples for attaining the higher accuracy.The simulation results show that the proposed method can identify 10 types of power quality disturbances including two complex disturbances,and have good noise immunity.And the higher accuracy can be achieved with less training and testing time compared with the stochastic gradient descent back-propagation (SGBP),least square support vector machine (LSSVM) and online sequential extreme learning machine method (OSELM).The proposed method is suitable for the online application.%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.
展开▼
