With the increasing complexity of grid structures

the distribution network has become a necessary means to improve grid scheduling efficiency

reliability

and economy. However

traditional distributed convex optimization algorithms are gradually losing the ability to meet the efficient operation requirements. In this context

machine learning has gained increasing attention. But the requirement for a large number of labeled samples has limited the practical application of machine learning. Aiming at this limitation

this paper proposes a self-supervised learning algorithm integrating the alternating direction method of multipliers(ADMM) to solve the distributed optimal power flow for a distributed network. First

two independent neural networks are constructed to simulate the interactive iteration of global variables and Lagrange multipliers in the ADMM

achieving mutual supervision between networks. The primal network is used to estimate global variables

while the dual network is employed to estimate Lagrange multipliers. Then

this paper proposes a loss function considering consistency constraints to guide the training of the primal network and ensure the convergence of the training process. Third

the convergence of the proposed method is analyzed and proven based on the Karush-Kuhn-Tucker (KKT) conditions

and the upper bound of the residual is derived. Finally

this paper conducts simulation experiments on the modified IEEE 123 bus system and the actual 276 bus distribution network in Suzhou to verify the effectiveness of the proposed method.