A Machine Learning Approach to Distinguish Faults and Cyberattacks in Smart Buildings

Advancements in smart control and the growing popularity of IoT devices have raised the demand of smart infrastructure. Smart buildings play a significant role in establishing a smart infrastructure. These buildings compose different subsystems that are connected through a complex cyber-physical network to a building management system (BMS). BMS monitors and controls components of a building with the help of various electronic components such as sensors, actuators, and measurement units. These devices are connected to the control center through communication networks. Due to the presence of a communication network, the system is always vulnerable to attacks in which an attacker can manipulate the sensors and measuring units and inject false data into the system. State estimation based techniques are used for fault detection, however, these techniques have limitations that encourage the demand for a more reliable system. The paper proposes a machine-learning algorithm to distinguish between normal state, fault state, and attack. The classification between fault and attack is achieved using the classifier learner tool in MATLAB. Support vector machine algorithm (SVM) is used to classify and through a comparative study, its superiority over other machine learning algorithms is shown. To validate the results various scenarios are considered to classify the different operating conditions.