Comparison of Machine Learning Based Anomaly Detection for Energy Consumption Values in SDN-IoT Based Home Area Networks

The problems of traditional electricity grids have led to the emergence of smart grids. Unlike traditional energy systems, smart grids play an important role in the energy sector with their flexibility, programmability and reliability. However, the heterogeneous structure of smart grids consisting of different devices and protocols poses some problems in terms of complexity, service quality and security. In the literature, SDN (Software Defined Networks) paradigm is proposed as a solution to these problems. SDN and smart grid integration makes the energy sector more efficient, reliable and sustainable. On the other hand, smart meters used in the consumption area of smart grids provide instantaneous transmission of energy production and consumption data in homes to the center. With the support of IoT (Internet of Things) of these meters and components in the home area network (oven, IP camera, TV, etc.), the energy supply and demand balance can be managed more smoothly. In this study, a software-defined and IoT-based smart home architecture is proposed to obtain real energy consumption data. The proposed architecture is developed and implemented on the Mininet simulator with python code. As a result of simulations run under different process and attack scenarios, energy consumption data sets were created. A comparison of the anomaly detection performances of machine learning algorithms on the data sets that are considered to contribute to the literature has been made. As a result of this comparison, it was observed that the success rate of the random forest algorithm was higher than the other algorithms with 90-95 percent.