The increasing penetration of the power grid with renewable distributed generation causes significant voltage fluctuations. Providing reactive power helps balancing the voltage in the grid. This paper proposes a novel adaptive volt-var control algorithm on the basis of deep reinforcement learning. The learning agent is an online-learning deep deterministic policy gradient that is applicable under real-time conditions in smart inverters for reactive power management. The algorithm only uses input data from the grid connection point of the inverter itself; thus, no additional communication devices are needed and it can be applied individually to any inverter in the grid. The proposed volt-var control is successfully simulated at various grid connection points in a 21-bus low-voltage distribution test feeder. The resulting voltage behavior is analyzed and a systematic voltage reduction is observed both in a static grid environment and a dynamic environment. The proposed algorithm enables flexible adaption to changing environments through continuous exploration during the learning process and, thus, contributes to a decentralized, automated voltage control in future power grids.

Vanadium redox flow batteries (VRFB) suffer from capacity fades owing to side reactions and crossover effects through the membrane. These processes lead to a deviation of the optimal initial average oxidation state (AOS=+3.5) of vanadium species in both half‐cell electrolytes. To rebalance the electrolyte solutions, it is first necessary to determine the current AOS. In this study, a new method was developed that enables an accurate determination of the AOS. A potential‐step analysis was performed with mixed electrolyte solutions of both half‐cells during the initial charging. The potential was recorded with a simple open‐circuit voltage (OCV) cell, and the potential‐steps were analyzed. A correlation between the duration of the potential plateaus in the OCV and the amount of vanadium ions of a certain oxidation state in the half‐cell electrolytes was found and used to precisely determine the AOS with a maximum error of 3.6 %.