基于改进GMM-CNN-GRU混合的非侵入式负荷监测方法研究

为挖掘用户侧节能减排潜力，对用户用电行为进行精细化分析和管理，提升电能利用效率，提出了一种基于高斯混合模型聚类和深度神经网络相结合的非侵入式负荷监测方法。首先，针对同一电器常出现功率相近但运行状态不一致问题，利用高斯混合模型聚类算法中“软分类”和类簇灵活的优势，对负荷工作状态进行精细分类，形成负荷用电设备实际运行情况的负荷状态特征库。其次，针对常见的应用于非侵入式负荷监测模型的深度神经网络在多标签分类时存在识别精度低等问题，提出卷积神经网络与门控循环单元混合的深度神经网络模型。最后，综合考虑外部环境数据对家庭用户用能习惯的影响，在AMPds2数据集上开展验证分析，并与其他模型进行对比。结果表明，所提的非侵入式负荷监测模型具有较高的准确性。

Non-invasive load monitoring based on an improved GMM-CNN-GRU combination

A non-intrusive load monitoring method based on Gaussian mixture model clustering combined with a deep neural network is proposed to explore the potential of energy saving and emission reduction at the customer side. It also fine-tunes the analysis and management of customer electricity consumption behavior, and improves the efficiency of electricity use. First, we tackle the problem that the same electrical appliance often has similar power but inconsistent operating status. In order to classify the load working status in fine manner, the advantages of "soft classification" and flexible clustering in the Gaussian mixture model clustering algorithm can be used to form a load status feature library that conforms to the actual operating conditions of electrical equipment. Secondly, note that in the common deep neural networks applied to non-invasive load monitoring models, there are problems such as low recognition accuracy in multi-label classification. Thus a deep neural network model with a mixture of convolutional neural networks and gated recurrent units is proposed. Finally, the validation analysis is carried out on the AMPds2 dataset by considering the influence of external environmental data on the energy consumption habits of household users, and the results are compared with other models. The results show that the proposed non-invasive load monitoring model has high accuracy.This work is supported by the National Natural Science Foundation of China (No. 61872230).