Decoupling prediction of cooling load and optimizing control for dedicated outdoor air systems by using a hybrid artificial neural network method

Abstract

Dedicated outdoor air systems (DOAS) can utilize high cooling water temperatures to achieve independent temperature and humidity control, which improves the energy efficiency of the system. Although many studies have investigated the energy consumption of DOAS under conventional controls, there is a lack of a cooling load (sensible load and latent load) decoupling control method to optimize DOAS operation. To address this challenge, this study introduces an Attention-Convolutional neural network-Long short-term memory (ACL) model, a hybrid deep learning framework explicitly designed for DOAS cooling load prediction. Unlike traditional approaches, the proposed ACL model decouples sensible and latent cooling loads, enabling precise load forecasting. First, a convolutional neural network (CNN) extracts critical cooling load features from building datasets. Finally, the ACL model-based control strategy is implemented through a co-simulation framework to optimize DOAS operating parameters. The results show demonstrate that the ACL model achieves an average prediction error of 5.7 %, with mean absolute proportional errors of 2.8 % for sensible cooling load and 1.9 % for latent cooling load. Moreover, the optimized ACL control strategy reduces DOAS power consumption by 7.7 %, ensuring energy-efficient operation in high-temperature, high-humidity environments. This study provides a new cooling load decoupling prediction control approach for DOAS, offering substantial energy savings.