Data-Driven Cooling Energy Consumption Model for High-Rise Buildings in Hot and Arid Climates

Planning, development, and operation of District Cooling Systems (DCS) require an urban scale modeling of the cooling load that the system is going to supply. An accurate cooling load estimation is crucial to creating an efficient, sustainable, and resilient operation. It is even more important for hot and arid climates as the overestimation of cooling loads is correlated with excessive capital and operation costs. In this regard, this study presents the development of an archetype-based data-driven building cooling load model for different high-rise building archetypes using a case study of the Marina district of Lusail City, Qatar. First, representative building archetypes are developed using EnergyPlus software to model the typical energy consumption behavior of the most prevalent buildings in the region. Second, a parametric simulation is performed automatically using RStudio to analyze the effect of varying parameters on building cooling load, as well as to obtain an artificial dataset covering different building configuration cases. The output dataset is ultimately used for the training and testing of three regression models: Multivariate Linear Regression (MLR), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGB). The results predicted by the models are then benchmarked against unseen cases developed with EnergyPlus software and the performance of the models is compared. According to the results, MLP and the XGB models yield excellent predictive capacity, which is manifested by a high R2 score on the training and testing sets, as well as on the unseen cases. The XGB model outperforms the MLP and MLR, exhibiting more accurate predictions of cooling energy consumption for the three building energy models studied in this work. On the other hand, MLR provides high spatial and temporal resolutions at sufficient accuracy with fast prediction capability at a relatively low computational cost compared to physical simulations.