UC Irvine

Traditional fossil-fuel-based energy sources are limited and have significant environmental impacts, contributing to climate change. Reducing global carbon dioxide (CO2) emissions to net zero by 2050 is essential to limit global temperature rise to 1.5 °C. Building operations account for 30% of global energy consumption and 20% of US energy-related greenhouse gas (GHG) emissions, necessitating effective strategies for managing building energy. This thesis aims to develop advanced machine learning models for forecasting building energy consumption using data from the University of California, Irvine (UCI) campus. Traditional models often struggle with complex datasets, while machine learning models like Artificial Neural Networks (ANN) and Recurrent Neural Networks (RNN) offer superior analysis and interpretation capabilities. Optimizing training functions and parameters is crucial for improving machine learning efficiency and accuracy. Enhancing model interpretability and performance is essential for practical application, especially for new systems with limited data. Three models were investigated in this work: two ANN models using the Levenberg-Marquardt and Bayesian Regularization algorithms, and a Long Short-Term Memory (LSTM) model. Early stopping criteria were evaluated for the ANN models to avoid overfitting, and different validation test failure values were assessed for forecast accuracy. The LSTM model's input parameters, such as timesteps and batch sizes, were configured for optimal training. The RNN model's input parameters, such as timesteps and batch sizes, were configured for optimal training. The models' forecasting abilities were compared for short-term (24 hours) and long-term (336 hours) predictions. Both ANN models demonstrated strong forecasting abilities, while the RNN model effectively captured daily building load trends but required further refinement. All three models showed potential in learning hourly time series data and are applicable beyond building energy forecasting, including fields like finance, healthcare, and weather forecasting.