基于多尺度自适应残差卷积神经网络的新能源配电网故障定位技术

Fault location technology for new energy distribution network based on multiscale adaptive residual convolutional neural network

随着新型电力系统建设的推进,分布式新能源接入配电网的比例不断提升,使得电网潮流分布更加复杂,对配电网的故障定位要求越来越高,导致现有的故障定位方法准确率低、稳定性差。对此提出一种基于多尺度自适应残差卷积神经网络的新能源配电网故障定位方法。首先,对故障电流使用变分模态分解,得到一系列征模态函数;然后,使用多尺度自适应卷积动态调整卷积核尺寸、残差卷积提升网络学习能力的方式构建多尺度自适应残差卷积神经网络模型,特征学习输入的故障电流本征模态函数;最后,经过Softmax分类器实现故障区段分类,完成故障定位。仿真结果表明,所提方法面对新能源接入的配电网能够实现不同故障的准确定位,并且对高阻接地故障仍然具有较高的准确率。和常见的卷积神经网络、支持向量机相比,配电网故障定位准确率分别提升了5.63%、9.31%,验证了该方法的有效性。

With the advancement of the construction of new power systems,the proportion of distributed new energy connected to the distribution network continues to increase,making the distribution of power flow more complex and requiring higher requirements for fault location in the distribution network,resulting in low accuracy and poor stability of existing fault location method.A fault location method for new energy distribution networks based on multi-scale adaptive residual convolutional neural networks is proposed.Firstly,the fault current is decomposed using variational mode decomposition to obtain a series of eigenmode functions.Then,a multi-scale adaptive residual convolution neural network model is constructed by dynamically adjusting the size of the convolution kernel and enhancing the learning ability of the network through residual convolution.The feature learning input fault current eigenmode function is used.Finally,the Softmax classifier is used to classify fault segments and complete fault localization.The simulation result show that the proposed method can accurately locate different faults in the distribution network connected to new energy,and still has high accuracy for high resistance grounding faults.Compared with common convolutional neural networks and support vector machines,the accuracy of fault location in distribution networks has been improved by 5.63%and 9.31%,respectively,verifying the effectiveness of this method.