Anti-intrusion method for low-voltage power line carrier communication system

Abstract

The invention discloses an anti-intrusion method for a low-voltage power line carrier communication system, which comprises the following steps of: preprocessing data; judging feature selection convergence, wherein if convergence is judged, online intrusion detection is carried out, and otherwise, the next step is carried out; feature selection improvement based on bidirectional search; learning device improvement based on a neural network algorithm; and using a learning device for training convergence judgment, wherein if convergence is judged, feature selection convergence judgment is carried out, otherwise, the previous step is repeated, optimization is carried out on an unbalanced data set by adopting a FocalLoss loss function, the anti-intrusion probability and intrusion detection precision of the low-voltage power line communication network are effectively improved, a CNN-Focal classification model is provided. According to the model, threshold convolution and Softmax in the convolutional neural network are applied to the field of intrusion detection for multi-classification, the problems of low calculation efficiency, complex calculation process and poor convergence of a package type feature selection mode are solved, and the accuracy and calculation efficiency of intrusion detection are improved.

Description

technical field [0001] The invention relates to the technical field of carrier communication, in particular to an anti-intrusion method for a low-voltage power line carrier communication system. Background technique [0002] At present, most of the intrusion detection systems at home and abroad are used for wireless network intrusion detection, which is not only slow in detection speed but also prone to false alarms and poor real-time performance. Coupled with the particularity of low-voltage power line communication, the security of low-voltage power line communication networks cannot be guaranteed. In recent years, methods such as injection clustering, association rule method, data mining and anomaly mining have appeared successively, but it is difficult to achieve the expected effect in today's complex power line communication network, and cannot meet the online intrusion detection system of modern power line carrier communication network. Improving the anti-intrusion lev...

AI Technical Summary

Problems solved by technology

[0002] At present, most of the intrusion detection systems at home and abroad are used for wireless network intrusion detection, which is not only slow in detection speed but also prone to false alarms and poor real-time performance. Coupled with the particularity of low-voltage power line communication, the security of low-voltage power line communication networks cannot be guaranteed. In recent years, methods such as injection clustering, association rule method, data mining and anomaly mining have appeared successively, but it is difficult to achieve the expected effect in today's complex power line communication network, and cannot meet the online intrusion detection system of modern power line carrier communication network. Improving the anti-intrusion level of low-voltage power line carrier communication and the ability of rapid and real-time detection of intrusion has become an urgent problem to be solved in the power line carrier network intrusion detection scheme.

[0003] In recent years, deep learning has achieved good results in the fields of speech recognition, image recognition, and natural language processing. Deep learning can extract abstract high-level features from original features, and does not need to select features based on expert experience. Because of its powerful learning Scholars at home and abroad have tried to apply deep learning technology to the field of network security. Although the above method has achieved good results, it only uses the official training set in model training and testing, which has certain limitations. , although the existing methods have a high anti-intrusion rate, due to the large dimension of the data, especially in the data processing stage, feature extraction will be performed based on expert experience, which not only takes a long time to select appropriate features, but may also damage the Due to the correlation between data, some effective features are missed, and because of the inherent low calculation efficiency, complex calculation process and poor convergence of the packaged feature selection mode, the existing optimization methods are difficult to achieve in practical applications. For real-time data of a huge power communication network, it is often easy to fall into the curse of dimensionality problem, which limits the accuracy and computational efficiency of intrusion detection

Images