Neural Network Analysis of PLC Traffic in Smart City Street Lighting Network

Abstract

In the article, we present a system allowing to detect different kinds of anomalies/attacks in street lighting critical infrastructure realized by means of Power Line Communication. There is proposed a two-phase method of anomaly detection. Firstly, all the outliers are detected and eliminated from the examined network traffic parameters with the use of the Cook’s distance. Data prepared this way is used in step number two to create models based on multi-layer perceptron neural network and autoregressive neural network which describe variability of the examined street lighting network parameters. In the following stage, resemblance between the expected network traffic and its real variability are analysed in order to identify abnormal behaviour which could demonstrate an attempt of an anomaly or attack. Additionally, we propose a recurrent learning procedure for the exploited neural networks in case there occur significant fluctuations in the real Power Line Communication traffic. The experimental results confirm that the presented solutions are both efficient and flexible.