IoT-Enabled Energy Optimization in Smart Buildings: A Machine Learning Approach to Predicting and Explaining Energy Consumption

Abstract

Much of the world's energy goes into buildings, and the spread of Internet of Things (IoT) sensor networks opens up practical avenues for managing that consumption through data. Using simulated IoT sensor data from a representative multi-zone commercial smart building, this study evaluated how total building energy consumption responded to occupancy, indoor temperature, relative humidity, lighting load, and heating, ventilation, and air-conditioning (HVAC) operation. To test predictions of hourly energy consumption, three machine learning algorithms were built and assessed against one another: Random Forest, Gradient Boosting, and an Artificial Neural Network. We assessed model performance through the Mean Absolute Error, the Root Mean Square Error, and the coefficient of determination. Next, a sensitivity analysis pinpointed which environmental and operational variables had the strongest impact on predicted energy use. Gradient Boosting yielded the lowest prediction error and highest coefficient of determination, with Random Forest running a close second; the Artificial Neural Network, by contrast, needed substantially more tuning before reaching similar accuracy. According to the sensitivity analysis, HVAC operational status and occupancy level drove energy consumption more than any other factor, with indoor temperature coming next. These results show that combining ensemble tree-based algorithms with live IoT sensor streams can deliver dependable, near real-time energy forecasts and help guide occupancy-responsive control schemes. We discuss what these results mean for facility managers, building automation designers, and policymakers working to cut operating energy costs and carbon emissions, and we suggest directions for future work.