A Distributed Online Learning Approach for Energy Management With Communication Noises

An online distributed neurodynamic optimization method for energy management in smart grids is proposed to minimize the operational cost of smart grids, considering various constraints, such as output bounds, state of charge of battery storage systems, network congestion and voltage limits. The proposed distributed neurodynamic optimization method adopts a consensus protocol using the signs of the information differences between neighbors for information exchanging and a recurrent neutral network for optimization without dual multipliers. It has the following merits. First, the calculation and information exchange of dual multipliers are avoided to mitigate the computational complexity and reduce communication bandwidth. Second, the proposed optimization method is robust against communication noises from various components. Third, the uncertainties of renewable energy and loads are addressed by the online optimization approach, which takes advantage of the historical data. The convergence, optimality, consensus and robustness of the proposed distributed neurodynamic optimization method are rigorously proved in theory and verified in case studies on the modified IEEE 33-bus and IEEE 123-bus distribution systems.