为了克服已有的基于神经网络的谐波测量方法，在采用BP算法训练时易陷入局部最小和易出现过拟合现象等神经网络固有缺陷，提出了一种将支持向量回归（SVR）应用于谐波测量的方法：该方法基于模拟并行谐波测量装置的基本原理，即从频域的观点，任何非正弦周期波形经过傅立叶级数展开，可以看成是由基波和各高次谐波迭加而成，利用SVR来实现模拟并行谐波测量装置中带通滤波器和检波器的功能。根据电力谐波的特点，从理论上构造训练数据，对该SVR模型进行训练，该模型输入为待测量信号，即在一个周期内的采样值，输出为待测的各次谐波幅值。大量仿真表明，提出的SVR谐波测量方法具有较好的测量精度。 为了扩大SVR谐波测量模型的测量范围和测量精度，本文将前人提出的合理压缩训练样本的方法应用于SVR谐波测量的建模中，构成了一个完整的谐波测量方法。该测量方法在每次测量时，首先利用离散的谐波采样值测出初相角，然后选择合适的SVR模型进行谐波测量。大量仿真验证了该方法的有效性。 综合上述方法，在MATLAB环境编写了完整的谐波测量程序，进行了大量的仿真研究，并与应用BP算法训练的神经网络的方法进行了比较。通过仿真实验，表明了本文建立的SVR谐波测量模型具有较好的测量精度和容噪能力，具有较强的稳健性。

Some disadvantages of the existing approach for measuring harmonic based on Artificial Neural Network methods are pointed out. The SVR is applied to harmonic measurement in this paper due to its obvious advantage such as good generalization ability ,unique and globally optimal solutions: Using the basic principle of analog parallel harmonics measurement device, some SVR modern for measuring harmonic are built. According to the primary characters of harmonics in power system , the training samples are made based on principle analyzing. Large numbers of simulation results illustrate the presented approach could acquire preferable measure precision. To increase the measure extension and measure precision, the existing approach for reducing training samples is used in building the SVR moderns: At first, the harmonic phase is determined, then the appropriate SVR moderns to be used for measuring the harmonic amplitude. Simulation illustrates the effectiveness of this method. In addition, an integrated measuring harmonic program is completed in MATLAB based on above methods and the BP Network is built also. Large numbers of simulation data are used to test the program. Simulation illustrates the capability of tolerating noise and robustness characteristics of this method.