Double Deep Q-learning Based Approach for Power Flow Optimization in Distribution Networks

With the increasing concerns on the environment and fossil resources, more and more renewable energies and energy storage systems (ESS) are embedded in active distribution networks (ADN). The stochastic nature of renewable energies brings great challenges to the power system operation. In this context, a double deep Q-learning based method is proposed to optimize the power flow of ADN considering the randomness of the load demand, wind speed. Experiment results validate that the proposed approach can master the optimal management strategy of ADN under constrains. Comparison results demonstrate that the proposed data-driven method can achieve better results when dealing with the uncertainty environment.