Novel Data-Driven Distributed Learning Framework for Solving AC Power Flow for Large Interconnected Systems

Abstract

Recent advancement in power systems induces complexity in large-scale interconnected systems and poses challenges in performing security assessment studies at various operating conditions. Traditional model-based methods are computationally intensive and may not meet the requirements for real-time applications. This paper presents a novel data-driven framework for accelerating the process of obtaining multiple AC power flow (ACPF) solutions for large systems using deep convolutional neural networks (DCNN). DCNN models are designed and trained using various representative power flow cases from a system, whose outputs can be used to perform steady-state security assessment studies. Distributed training with multiple Graphical processing units (GPU)s is implemented using TensorFlow to reduce computation time. The proposed framework is implemented to identify critical buses and recognize the ACPF cases expected to cause steady-state bus voltage violations. The efficacy and feasibility of the proposed framework are evaluated on the Western Electricity Coordinating Council (WECC) 2028 system. Results demonstrate that the proposed framework is highly accurate and possess good interpretability in performing various transmission planning and operation assessment studies for large scale power networks.

Keywords

• Contingency screening
• security assessment
• deep convolutional neural networks (DCNNs)
• feature selection
• machine learning (ML)
• deep learning (DL)