基于堆栈稀疏降噪自编码的电压暂降源识别方法

A Voltage Sag Sources Identification Method Based on Stacked Sparse De-noising Auto-Encoder

为有效识别电压暂降源,文章提出一种基于堆栈稀疏降噪自编码的电压暂降源识别方法,用于识别单一和复合电压暂降源。稀疏降噪自编码网络是在自编码网络的基础上加入稀疏性限制,同时在输入信号中加入按一定概率分布的噪声构成的深度神经网络,将其逐层堆砌形成堆栈稀疏降噪自编码网络。实验首先利用无标签训练集初始化网络中的权重和偏置项,再利用有标签训练集进行一次有监督的微调,使网络能够学习输入信号中更深层次的特征,最后采用SoftMax函数对特征进行分类。结果证明,该方法对电压暂降源的识别率高,同时,基于实测数据增量训练下的模型将具有更好的泛化能力,能够很好地应用于实际工程项目中。

In order to effectively identify the voltage sag sources,this paper presents a voltage sag sources identification method based on Stacked Sparse De-noising Auto-Encoder (SSDAE),which is used to identify single and composite voltage sag sources.The Sparse De-noising Auto-Encoder network (SDAE) is based on the Auto-Encoder network,adds sparsity restrictions and a certain probability distribution of noise to the input signals,then stacks them layer by layer to form SSDAE network.The experiment first uses unlabeled data sets to initialize the weights and offset of the network,and then uses labeled training sets to perform a supervised fine-tuning so that the network can learn the deeper features of the input signals.Finally,the SoftMax function is used to classify the features.The result shows that this method has high recognition rate for voltage sag sources,at the same time,the model based on incremental training of measured data will have better generalization ability and can be well applied to the actual engineering projects.