Research on Attack Detection Based on Random Forest Improved Algorithm for Power Line Carrier Communication.

With the rapid development of distribution network, more and more intelligent devices are connected to power network through distribution communication network. However, with the increasing of the types and quantity of network attacks, the traditional attack detection methods not only have more and more prominent problems in missing and false detection, but also have the deficiency of low information utilization, which can not meet the security requirements of distribution services for distribution communication network. The random forest algorithm has the advantages of good classification accuracy and no need to preprocess data. Random forest is chosen as the attack detection algorithm in this paper. Aiming at the problem that the model construction of random forest is time-consuming and the classification accuracy of small proportion category under unbalanced samples is low, an improved random forest algorithm and an improved random forest algorithm combined with ReliefF improved algorithm are proposed to improve the classification accuracy of small proportion category.