Graph neural network-based spatio-temporal methodology for hydraulic-thermal modelling of district cooling systems

Hydraulic-thermal modelling plays a crucial role in effective management of district cooling systems, enabling fault detection and diagnosis and pump speed control to achieve energy savings. The development of traditional detailed physical hydraulic models is time consuming and labour intensive, and the models developed need to be calibrated several times during the operational stage. Data-driven modelling has made significant strides in capturing temporal relationships using advanced machine learning algorithms. However, existing data-driven modelling methods often make simplistic assumptions about water distribution and generally overlook structural relationships, such as the topology of cooling water pipes and pumps. To address this issue, this study presents a novel spatio-temporal data-driven methodology for representing district cooling systems as graphs and modelling the hydraulic system. A case study of a district cooling system based on real cooling load simulation is carried out to evaluate the efficacy of the proposed methodology. Graph neural network-based models that consist of a graph layer and a recurrent layer were developed to capture the structural and temporal relationships separately. The graph layer utilizes graph convolutional network (GCN) and graph attention network (GAT) to learn the structural relationships from the input graphs. The recurrent layer utilizes Long Short-Term Memory (LSTM) to learn the temporal relationships from massive historical operational data. The developed models generated better prediction performance than conventional deep learning models. This methodology provides a new approach to analyzing hydraulic systems that can facilitate effective management of district cooling systems. By incorporating both the structural and temporal relationships in the data-driven modelling process, the methodology offers a more comprehensive understanding of the hydraulic system and can provide valuable insights for optimizing the system's energy efficiency and fault detection. It has the potential to be widely applied in the field of district cooling systems and other related hydraulic systems.