基于三维卷积神经网络的配电物联网异常辨识方法

The power grid and communication network are highly coupled in the distribution Internet of Things (DIoT). The anomalies of a single network interact with another network, which may lead to the expansion of anomaly range. However, it is difficult to comprehensively and accurately identify the types and locations of the anomaly source in the DIoT by using the information of the power grid or the communication network alone. Therefore, this paper proposes an anomaly type identification and location method for DIoT based on the three-dimensional convolutional neural network (3D-CNN). Firstly, the communication flow characteristics of DIoT are analyzed and an interactive simulation model of DIoT based on Simulink and OPNET is built. Secondly, a 3D-CNN oriented sample construction method is proposed, in which the electric parameters and communication flow information of each node in DIoT are composed into a feature sub-pixel, and the state of DIoT at each moment is represented as a feature frame, thus forming a cubic sample matrix that contains the anomaly process of DIoT. Thirdly, a deep learning model is built, which includes the three-dimensional feature extraction network and the hierarchical softmax classifier. By extracting and identifying the abnormal information hidden in the cubic sample matrix, the type and location of anomalies in DIoT could be determined simultaneously. Finally, the model is tested by using abnormal data of the IEEE 33-node DIoT. The results show that the proposed method can precisely classify and locate the short-circuit fault, communication interruption fault, and protection maloperation and rejection caused by abnormal communication data.