Abstract:

With the widespread use of internet technology in smart grids, it is particularly important to identify intrusion attacks in power systems. Based on the communication network architecture in Advanced Metering Infrastructure（AMI）, an AMI communication intrusion detection scheme combining K Nearest Neighbor （KNN） and optimized feature engineering is proposed in response to the smart grid intrusion detection requirements. Intrusion attack flow can be identified through four modules： data collection, data preprocessing, feature engineering and model training. In feature engineering module, the features inputted into KNN training model are optimized by text feature extraction method, and the redundant feature vectors are removed based on the information gain values. In model training part, the types of data are judged by the labels of the k nearest neighbour training samples. The proposed scheme was tested on public intrusion detection data sets ADFA-LD, and the detection accuracy of various intrusion attacks was obtained. The experimental results show that the detection performance of this scheme is superior to the traditional intrusion detection scheme, with an 21.96% increase in the classification accuracy under the optimal feature extraction model.

Key words: smart grid, intrusion detection, Advanced Metering Infrastructure, KNN, feature engineering, ADFA-LD, Energy Internet, power information security