CH-24-C014 - Data-Driven Modeling of IoT-based Smart Buildings for Energy Prediction

Building environments are highly stochastic and influenced by outdoor weather conditions, operating schedules, internal thermal gains, and occupant behavior, to name a few. Traditional physics-based models require elaborate inputs specifying detailed building designs and air-flow systems, while statistical models are often less accurate and limited to the facility and conditions under which the models were constructed. Both types of models are context-specific and time-consuming, making them unsuitable for supporting energy analysis or intelligent control of complex systems with massive amounts of noisy data. With the increasing application of Internet of Things (IoT)-based sensing and controls in buildings, large volumes of building data have become available in the cloud. This highlight motivates the urgent need for robust data-driven models using machine learning (ML) techniques to facilitate energy analytics and prediction in IoT-based smart buildings. The primary objective of the study is to develop a robust data-driven model for building energy analytics and optimization. To achieve this goal, a case study was performed on an operational IoT-based building with real-time sensed data. Models based on Support Vector Machine (SVM), Feedforward Neural Networks (FNN), and Recurrent Neural Network (RNN) were developed to relate highthroughput streaming data to observable parameters of interest, learn from the operational patterns of the building response, and make predictions about future energy consumption. An automated ML pipeline was established consisting of streaming data collection from the cloud, data preprocessing, feature selection, feature engineering, model training, and validation. The model was specifically used to forecast the building energy performance. The data-driven model has been proven to be highly accurate in predicting energy consumption, thereby enabling optimal energy consumption and enhancing the overall performance of the building.