基于多层多核卷积神经网络的非侵入式负荷监测方法研究

Research on Non-Intrusive Load Monitoring Method Based on Multi-Layer and Multi-Kernel Convolution Neural Network

非侵入式负荷监测仅依靠测量得到的总负荷的电压、电流与功率等承载电力信息的信号就实现负荷监测,无需额外的计量装置和线路改造,因此得到广泛研究。针对传统深度神经网络分解模型准确度仍不能满足实际需求的现状,提出了一种基于多层多核卷积深度神经网络分解模型。为体现不同设备的特性,模型在数据分割时采用不同的序列长度。然后,模型将分割后的数据先通过高维映射,将输入的功率时间序列映射到高维向量,再利用多层卷积法与多核卷积法共同构建出的深度神经网络对生成的信息向量进行特征提取,经多次迭代学习生成负荷分解模型。与多种用于非侵入式负荷分解的深度学习方法相比,本模型对负荷识别准确率提升效果显著,在REDD数据集上的识别准确率达到99.41%。

Only by measuring the total load voltage, current, power and other signals carrying power information, non-invasive load monitoring can achieve load monitoring without additional metering devices and line transformation, so it has been widely studied. Aiming at the low accuracy of the traditional deep neural network decomposition model, a decomposition model based on multi-layer and multi-core convolution deep neural networks is proposed. In order to reflect the characteristics of different devices, different sequence lengths are used in data segmentation. Then, the model first transforms the input power time series into high-dimensional vectors by high-dimensional mapping and then designs the multi-layer and multi-core convolution deep neural networks to extract the features of the generated vectors. Finally, the load decomposition model is generated after repeated iterative learning. Compared with other deep learning methods for non-intrusive load decomposition, the model improves the accuracy of load identification obviously, and the recognition accuracy on REDD data set reaches 99.41%.