Power System Operation Mode Adjustment based on GNN Broadcasting and Gradient Descent

The operation mode arrangement is one of the critical components in practical power grid operations. Operators must design a safe and stable operation mode by considering factors such as load levels, maintenance schedules, and equipment contingencies. However, with the increasing scale of power grids and the growing number of control elements, the traditional manual arrangement method based on experience struggles to efficiently establish a safe and stable operation mode. To address this challenge, this paper proposes an intelligent power system operation mode adjustment method. By training a physicsinformed Graph Neural Network (GNN), a numerical mapping from system states to power flow results is constructed. The proposed method leverages this mapping to adjust system states via gradient descent, thereby alleviating overloaded branches. Additionally, it integrates the N-1 criteria, which induce topology changes, into the adjustment process through the GNN broadcasting mechanism. The newly generated operation mode inherently satisfies the N-1 security verification. Simulations conducted on the IEEE 30-bus system demonstrate that the proposed method can rapidly generate an operation mode satisfying the $N-1$ verification without requiring extensive memory resources.