基于新型卷积神经网络的非侵入式负载监测方法

现有非侵入式负载监测技术处理负载规模变化的能力弱，且随着负载的种类复杂化与数量的增多，其具有估计精度不高的问题。建立了一种考虑用户用能多时间尺度耦合特性，并具有规模化处理能力的新型卷积神经网络，以提高复杂规模化负载估计的精确性。该神经网络包括多时间尺度感知与特征提取模块、自我关注模块和对抗损失模块等，多时间尺度感知与特征提取模块可获取与整合不同时间尺度负载数据的耦合特征，自我关注模块和对抗损失模块根据耦合特性来进一步提高监测模型的估计精度。最后，通过仿真分析验证了所提方法的有效性和优越性。

Non-invasive Load Monitoring Method Based on Novel Convolutional Neural Network

In view of the weak ability of existing Non-Intrusive Load Monitoring （NILM） technology to deal with the changes of load scale， and its estimation accuracy is not high with the increase of the types and quantities of loads. A multi-time scale coupling characteristic considering the user energy is set up. And a new convolutional neural network with scale processing ability is used to improve the accuracy of complex scale load estimation. The neural network includes multiple time scales perception and feature extraction module，self-care and against loss module. Multiple time scales perception and feature extraction module can obtain and integrate the coupling characteristics of load data with different time scales， focus on self and against loss module further improve the estimation precision of the monitor model based on the coupling characteristics. Finally，the simulation results verify the effectiveness and superiority of the proposed method.