A decentralized optimization framework for multi-MGs in distribution network considering parallel architecture

In view of centralized optimization facing the shortcomings of heavy communication burden, poor privacy, lack of autonomy or susceptibility to communication failures, a decentralized optimization framework is proposed to apply in parallel architecture particularly constituted by multi-microgrids (MMGs) in a distribution network based on accelerated analytical target cascading (ATC). Accelerated ATC algorithm can be applied to perform decentralized optimization in a sequential manner between distribution network agents and microgrid agents in a faster convergence speed compared with the traditional ATC algorithm. In order to ensure the stability of power flow and to incorporate unit commitment into the problem, a comprehensive economic dispatch model in the form of mixed-integer second-order cone programming (MISOCP) is developed and integrated into decentralized optimization framework. In our proposed decentralized optimization framework, each agent only needs to exchange a small amount of boundary information with its neighboring agents to find the feasible solutions without revealing their private operational information. The novelty of this work includes (1) the application and acceleration of the ATC algorithm in the proposed decentralized optimization framework; (2) the extensive investigation of the solution feasibility of the derived MISOCP problem for various penalty multipliers, scales and objective functions. Three types of microgrids (MGs) for residential, industry and business sectors are connected to the distribution network in parallel, respectively. Aiming at MMGs constructed by three MGs and six MGs in the distribution network, the proposed decentralized optimization framework is validated for the acceleration of ATC algorithm, varying penalty multipliers as well as different types of objective functions in terms of the feasibility of distributed solutions. Case studies based on the modified IEEE 33-bus distribution network are conducted to show the effectiveness of the decentralized optimization framework.