A Physics-Informed Neural Network Approach Towards Fast Gas Flow Calculation

Natural gas systems (NGS) and electric power systems are increasingly interconnected to enhance flexibility. However, the bidirectional energy flow problem, represented by partial differential algebraic equations (PDAE), remains a challenge for model-based solvers. Achieving feasible solutions with complete parameter availability in a large-scale system can take several minutes to hours. To address this, we propose a Physics-Informed Gas Flow Neural Network (PIGFNN) for gas flow calculations in NGS, effectively mitigating the time-consuming aspects of PDAE. The PIGFNN accurately captures heterogeneous gas states while ensuring a privacy-preserving interface. Additionally, we introduce a physics-informed loss function, NGSPILoss, which quantifies violations of dynamic constraints in NGS, ensuring that PDAE constraints are satisfied in gas flow calculations. Case studies demonstrate the effectiveness of the proposed method.