Learning-accelerated asynchronous decentralized optimization for integrated transmission and distribution systems over lossy networks

Abstract

With the rapid growth of distributed renewable energy in distribution networks, the synergistic operation between transmission and distribution networks is increasingly important. Decentralized optimization is employed in the coordination of integrated transmission and distribution networks to reduce the communication burden and protect regional information privacy. Existing works on decentralized optimization algorithms for integrated transmission and distribution networks have seldom addressed packet losses. Moreover, the synchronous operation requirement is difficult to achieve for many decentralized optimization algorithms to solve problems with computationally heterogeneous subproblems and during the occurrence of communication failure. In order to tackle these problems, this paper proposes a decentralized scheme based on the relaxed alternating direction method of multipliers for coordinating the operation of integrated transmission and distribution networks considering communication failures and asynchronous updates. A momentum–extrapolation prediction–correction method is designed to allow operators to forecast their neighbors’ missing boundary variable values and to be solved continuously. Simulation results show that compared with the classical alternating direction method of multipliers, the relaxed alternating direction method of multipliers can obtain the nearly same operation costs with fewer iterations. In addition, the proposed algorithm shows robustness against packet losses and asynchronous computations. It is also found that the prediction–correction technique makes the proposed algorithm converge faster compared with no prediction. • Step size can change the performance of proposed algorithms. • Packet losses are represented by a binary probabilistic distribution model. • Convergence rates are affected by packet losses and asynchronous updates. • A prediction-correction approach is used to forecast the missing information.