运用快速傅里叶变换(fast Fourier transform，FFT)进行电力谐波分析时存在频谱泄漏和栅栏效应，误差较大。在研究了窗函数与频谱泄漏的影响的基础上，提出一种改进型单输入多输出人工神经网络谐波分析算法。布莱克曼–纳托尔(Blackman-Nuttall)窗具有良好的旁瓣性能，能有效抑制频谱泄漏，基于布莱克曼–纳托尔窗的插值FFT算法可实现高准确度信号频率估计。单输入多输出神经网络模型根据信号频率估计值构建神经网络输入向量，获得准确的谐波幅值、相角值。给出了该算法用于电力系统谐波分析的算例。仿真结果表明：该方法能实现微弱幅值的谐波分量的准确分析，电力系统谐波检测准确度可满足实际测量需求。

The accuracy of the power system harmonic analysis by fast Fourier transform (FFT) is disturbed by the spectrum leakage and the picket fence effect. Considering the relationship between the window function and the spectrum leakage, this paper proposes an improved neural network with single input multiple output (SIMO) for harmonic analysis. The Blackman-Nuttall window has good side lobe behavior, which can reduce the spectrum leakage efficiently. Hence, the fundamental frequency was estimated by the Blackman- Nuttall window-based interpolation FFT algorithm, which also produced the input vectors of the SIMO neural network. By training the SIMO neural network, the harmonic parameters, including amplitudes and phases angles, were detected precisely. The simulation cases with the power system harmonic analysis were conducted. Simulation results show that the weak-amplitude harmonics can be detected from a signal with higher accuracy by using the proposed method, and the accuracy can completely meet the requirements of the harmonic analysis in power system.

国家自然科学基金项目(60872128)。