基于一维卷积神经网络的负荷分解方法

Load decomposition method based on one⁃dimensional convolution neural network

非侵入式负荷监测因其成本低、隐私性高,具有良好的应用场景。负荷分解方法是非侵入式负荷监测的主要技术难点之一,为提高负荷分解的精度,提出一种基于一维卷积神经网络的负荷分解方法。该方法首先以滑动窗口读取总负荷时间序列生成输入序列,解决深度学习模型不能输入长序列的问题;接着,以序列扩展模块自动提取输入序列的特征并重构为扩展序列,扩展了输入序列的特征信息;最后,采用端到点结构构建特征提取模块,提取扩展序列特征输出负荷分解结果,其中,序列扩展模块和特征提取模块共同构成了一维卷积神经网络模型。在公开的数据集UK_DALE上的实验结果表明,所提出的基于一维卷积神经网络的负荷分解方法具有可行性,与现有方法相比,该方法的分解性能更好,F_1得分更高。

The non⁃intrusive load monitoring(NILM)has good application prospect because of its low cost and high performance in keeping privacy.Load decomposition method is one of the main technical difficulties of non⁃intrusive load monitoring.Therefore,a load decomposition method based on one⁃dimensional convolution neural network(1D⁃CNN)is proposed to improve the accuracy of load decomposition.In the method,a sliding window is used to read the time series of total load,so as to generate an input sequence and solve the problem that the deep learning model is incapable for the input of the long sequences;the sequence expansion module is used to automatically extract the features of the input sequence and reconstruct it into an expanded sequence to expand the feature information of the input sequence;the sequence⁃to⁃point structure is used to construct a feature extraction module to extract the features of the extended sequence and output load decomposition results.The sequence expansion module and the feature extraction module together constitute a 1D⁃CNN model.The experimental results on the public data set UK_DALE show that the proposed load decomposition method based on 1D⁃CNN is feasible.In comparison with the existing methods,the proposed method has better load decomposition performance and higher F1 score.