Machine learning-based cooling load prediction and optimal control for mechanical ventilative cooling in high-rise buildings

Abstract

Ventilation has proved to be an effective solution for reducing building cooling load in high-rise buildings, i.e. ventilative cooling (VC), especially in cold climates. Mechanical ventilation system can achieve VC (i.e. mechanical VC) in high-rise buildings, but it needs appropriate control to reduce cooling related energy consumption and to consider the impact of climate change. This study aims to develop an optimal control method for mechanical VC and evaluate its energy performance, based on an advanced model. In this advanced model, a machine learning model was applied to predict building cooling load and energy models were developed to predict energy consumption. A case study was conducted on a real high-rise building located in Montreal (Canada). Using the measured data from the Building Automation System (BAS), the machine learning model, generated by an algorithm called Gradient tree boosting (GTB), was found to be the most accurate and was used in the optimal control method. The energy models that couple mechanical ventilation and chiller cooling were validated with the BAS measured data. Then, the optimal control method was applied to study the energy performance of mechanical VC under long-term climate conditions. The results indicate that the energy savings of mechanical VC will decrease around 16% during the summer but increase around 105% during the shoulder season due to climate change. The optimal nominal ventilation rate for mechanical VC in the typical meteorological year is twice of that in the 2080 s A1FI weather scenario.