Building HVAC Electric Load Demand Prediction: Balancing Learning Rate and Hidden Layers for Improved Model Performance

Abstract

This study examines the performance of a predictive model for building HVAC electric load demand under three distinct conditions. The analysis focuses on two key metrics: the coefficient of variation of the root mean square error (CVRMSE) and the coefficient of determination (R²). Results indicate a notable disparity in model fitting across the conditions. For conditions 1 (learning rate =0.0001, hidden layer =7) and 3 (learning rate =0.0001, hidden layer =5), an increase in iteration rounds leads to a decrease in CVRMSE, signifying enhanced prediction accuracy. Conversely, condition 2 (learning rate =0.01, hidden layer =7) exhibits an increase in CVRMSE with more iterations, suggesting reduced accuracy. The R² values consistently rise with additional iterations across all conditions, indicating improved model fit. However, condition 2 presents a slightly larger discrepancy between the training and test sets compared to conditions 1 and 3. These findings highlight the varying impacts of iteration on model performance across different scenarios. The study underscores the importance of tailoring model parameters, such as learning rate and hidden layers, to specific conditions to optimize predictive accuracy. This research contributes to the understanding of how iterative processes and model configurations affect the accuracy and reliability of HVAC load predictions, offering insights for future model development and application in energy management systems.