Intrusion Detection in Smart Buildings Using Energy Anomalies: A Long Short-Term Memory Model Approach

The increasing prevalence of smart buildings within urban environments necessitates advanced security measures to detect and mitigate potential threats. This study leverages the data by a private company ASHRAE, the ASHRAE - Great Energy Predictor III dataset (GEPIII). The research question is: How can anomalous energy consumption be used as a proxy for identifying intrusions in smart buildings? By establishing baseline energy consumption patterns for building operations, we investigate how deviations from these patterns may signal the presence of unauthorised individuals. The anomaly detection in this study focuses on deviations in energy consumption patterns, considering not only magnitude and frequency but also duration, timing, rate of change, consistency across similar conditions, correlation with external factors like weather, aggregate daily or monthly usage, geospatial distribution within the building, and statistical outliers. In this study, we employ a Long Short-Term Memory (LSTM) neural network for our anomaly detection task, capitalising on their ability to capture dependencies in sequential data. After training our LSTM model, we conducted extensive validation to assess its performance. The dataset provides meter readings from over 1300 commercial buildings, of which we used a subset of 100 randomly selected buildings for this study due to computational resource limitations. Using IoT with interconnected sensing devices in smart buildings to collect data, combined with AI is an emerging research area in building security. Results highlight the potential of this approach to provide tools for enhancing the security of smart buildings, with implications for broader urban safety systems. Broader implications are that threats can be detected pre-emptively by using the developed model, or buildings can be designed and then a simulation can be run against the developed AI model, influencing future building codes or policy changes for the governance of urban environments.