A graph neural network-based fault detection framework for combined building-integrated photovoltaics, energy storage, and building flexibility control systems

Abstract Fault detections among building-integrated photovoltaics (BIPV), battery energy storage (BES), and building energy flexibility (BEF) systems are essential during the phase of operation and maintenance. Despite the linkage among these three systems, most of the existing fault detection methods focused on individual systems, neglecting the interconnection between BIPV, BES, and BEF systems. These make detection results relatively isolated among these systems and lack reliability, which might cause additional labour and cost. This study presented a framework that illustrates the way of applying the graph neural network (GNN) to potentially detect failure and infer cause among BIPV, BES, and BEF systems based on an ontology established according to standards. The graph structure was built based on the topology structure of the existing ontology, and system operational data were input into corresponding nodes and edges. These mapped data were then pre-processed and sent into the GNN model, with the nodes and edges maintaining the built structure. After processing by the GNN model, the results of each node might be used to detect failures among these systems. The proposed framework offers valuable insights into the fault management practices within combined BIPV, BES, and BEF systems.