Integrating time series decomposition and multivariable approaches for enhanced cooling energy management

Predicting electricity consumption for building cooling is a critical challenge for improving sustainability, but existing simulation methods often overlook the interaction between system operations and energy use. This study presents a new approach that improves the accuracy of cooling electricity consumption predictions. The method combines time series decomposition with multivariable modelling, using Seasonal and Trend decomposition using Loess to disaggregate the demand data into trend, seasonal, and residual components. Multivariate regression models are then developed to predict the trend component, integrating a comprehensive set of system operating variables and environmental factors. This innovative modelling framework achieves an exceptional coefficient of determination of 0.9994 in the training set and 0.9905 in the validation set. By using trend component models, the important operating variables are effectively addressed, enabling optimization of the chiller system's operating variables. The minimum electricity consumption is assessed based on boundaries at the 80th, 90th, 95th and 99th percentiles of operating variables. The analysis revealed potential annual savings of 11.178 %–16.401 %, equivalent to 191,805 to 281,427 kW h, with distinct daily and seasonal variability patterns. The novel modelling approach empowers building operators to realize the drivers of energy use, enabling data-driven optimization strategies that enhance sustainability.