Prediction and Classification of Temperature Data in Smart Building using Dynamic Mode Decomposition

With the recent trends of smart cities, the development in the sector of Smart Buildings has emerged tremendously which consists of multiple layers coordinating and interacting with each other with the help of a building management system (BMS). This interaction of different layers in the smart building with the help of a communication channel leads to exposure of layers to vulnerabilities (cyber attacks) which may lead to anomalies condition. This kind of anomalies can be avoided by proper prediction of data and coordination among different layers of the building operation. However, to develop the model for prediction of data is quite time consuming and hence, the paper proposes the concept of Dynamic Mode Decomposition (DMD) for predicting data with help of past available data even in absence of system model. In this paper temperature profile of heating, ventilation, and air conditioning (HVAC) system in BMS is predicted with the help of available past data. Once the prediction of the temperature profile is achieved the machine learning algorithm is used to classify and identify the data as normal or anomalies condition. The two-fold contribution of the paper in the prediction of temperature using DMD where all system states may not be observable and classification of data using machine learning is validated considering different test scenarios and results show the effectiveness of the DMD method in the prediction of data as well as classification using a machine learning algorithm.