适应衰减直流分量估计的特定神经网络模型

A specific neural network for estimating decaying DC component

为了估计出曲线函数的参数,提出了一种特定神经网络应用于图形曲线的函数拟合,并针对故障电流信号中包含的指数衰减直流分量降低继电保护操作的精度和效率等问题,构造了一种适应衰减直流分量估计的神经网络模型(decaying DC neural network,DDCNN),推导出基于Levenberg-Marquardt算法的网络权值自适应学习方法.该模型与包含衰减直流分量故障电流信号模型的数学表达式一致,在迭代求解神经网络的权值后,可直接由权值估计出衰减直流分量的所有参数.实验仿真结果及对比分析表明,本文算法能获得比现有算法更高的估计精度,且其计算代价能够满足应用需求.

In this paper, a specific neural network is proposed to function fitting of the graph curve, and then all the parameters of the curve function are estimated. For the fault current signal usually contain a decaying DC component and this component may decrease the accuracy and efficiency of the operation of digital relay protection, a specific neural network model is constructed for estima ring decaying DC component, and then an adaptive learning algorithm based on Levenberg-Marquardt algorithm is derived. Due to the consistency of the mathematical formulas between the proposed neural network model and the fault currents containing decaying DC component model,parameters of the decaying DC component can be directly calculated using the converged weights of the proposed neural network. Simulation results and comparative analysis validate that this algorithm can achieve a higher accuracy than the state of the art algorithm. Moreover, the time cost of the proposed algorithm can be accepted for digital relay protection.