基于改进自编码网络的电力负荷识别

Power Load Identification Based on Improved Autoencoder Network

为提高电器设备识别的准确率,提出一种基于改进稀疏自编码网络和支持向量机算法相结合的电力负荷识别方法。以自编码网络的编码部分对输入数据的表征能力为基础,结合卷积神经网络特征提取的能力,采用卷积层替换自编码网络中的全连接层;通过在损失函数中加入惩罚项,进一步优化网络对负荷特征的提取能力;最后将特征放入粒子群优化后的支持向量机模型做识别。试验结果表明,方法能够有效地对电器进行多种类型的识别。相较传统自编码网络,改进后的模型泛化能力更强,识别率更高。

In order to improve the accuracy of electrical equipment identification,an electric load identification method based on the combination of improved sparse autoencoder network and support vector machine algorithm was proposed.Based on the ability of the coding part of the autoencoder network to represent the input data,combined with the feature extraction ability of the convolutional neural network,the convolutional layer was used to replace the fully connected layer in the autoencoder network;by adding a penalty term to the loss function,the ability of the network to extract load features was further optimized;finally,the features were put into the support vector machine model after particle swarm optimization for identification.The test results show that the method can effectively identify various types of electrical appliances.Compared with the traditional autoencoder network,the improved model has stronger generalization ability and higher recognition rate.