Machine learning for agile and self-adaptive congestion management in active distribution networks

Although congestion management via Demand Response (DR) has gain sufficient popularity recently, there are still some fundamental impediments to achieve a trade-off between demand flexibility scheduling and demand flexibility dispatch for congestion management. To find a solution to the challenge, the paper introduces the concept and design of an Agile Net, which is an agile control strategy for congestion management. The model of Agile Net has triple cores. First, it percepts the network environment by using the concept of demand elasticity. Second, it possesses an online model-free learning technique for the management of network externality, such as congestion. Third, it enables distributed system scalability. The efficiency of the proposed Agile Net is investigated by extending the simulation tool for DR paradigm for a generic low-voltage network of the Netherlands. Simulation results reveal a significant reduction in congestion over a year while confirming expected levels of performance.