Optimized Local Control for Active Distribution Grids using Machine Learning Techniques

Modern distribution system operators are facing a changing scenery due to the increasing penetration of distributed energy resources, introducing new challenges to system operation. In order to ensure secure system operation at a low cost, centralized and decentralized operational schemes are used to optimally dispatch these units. This paper proposes a decentralized, real-time, operation scheme for the optimal dispatch of distributed energy resources in the absence of extensive monitoring and communication infrastructure. This scheme uses an offline, centralized, optimal operation algorithm, with historical information, to generate a training dataset consisting of various operating conditions and corresponding distributed energy resources optimal decisions. Then, this dataset is used to design the individual local controllers for each unit with the use of machine learning techniques. The performance of the proposed method is tested on a low-voltage distribution network and is compared against centralized and existing decentralized methods.