基于线性神经网络的配电网非整次谐波检测溯源

Nonintegral harmonic detection and traceability of distribution network based on linear neural network

针对现有谐波检测装置对非整次谐波检测精度较低、检测结果有效性不足的问题,提出对线性神经网络进行改进,实现权值参数的灵活调整;并通过将BP神经网络与线性神经网络相结合,先进行基波检测再进行非整次谐波检测;接着引入加汉宁窗的快速傅里叶变换,进行数据预处理,以减少线性神经网络的计算量,进一步保证检测精度和运算速率。最后通过仿真验证改进后的非整次谐波检测模型的检测精度。结果表明,改进后的检测模型的检测误差极低,在10^(-4)的精度范围内,检测精度极高,且具有较高的抗噪能力,在随机噪声环境下的检测精度也存在明显提升,基本满足设计要求。

Aiming at the problems of low detection accuracy and insufficient effectiveness of existing harmonic detection devices for non integral harmonics,this paper proposes to improve the linear neural network to realize the flexible adjustment of weight parameters;By combining BP neural network with linear neural network,the fundamental wave is detected first,and then the non integral harmonic is detected;Then the fast Fourier transform with Hanning window is introduced for data preprocessing to reduce the amount of calculation of linear neural network and further ensure the detection accuracy and operation speed.Finally,the detection accuracy of the improved non integral harmonic detection model is verified by simulation.The results show that the detection error of the improved detection model is very low.In the accuracy range of 10^(-4),the detection accuracy is very high,and has high anti noise ability.The detection accuracy in random noise environment is also significantly improved,which basically meets the design requirements.