Optimal operation of grid-connected hydropower plants through voltage control methods


  • Emil G. Melfald
  • Thomas Øyvang




Optimal Control, Reinforcement Learning, Power loss models, Hydropower


The ongoing decarbonization, and the rapid increase in renewable penetration in the electric grid, will demand enhanced flexible operational schemes of the conventional hydropower plants. This paper explores a grid-connected hydropower plant’s best-efficiency operating conditions when meeting the renewable energy transition. Power loss models combined with various voltage control methods are investigated for achieving optimal operation. This simulation study is carried out on a static F13Single Machine Infinite Bus (SMIB) environment to perform voltage control comparisons. Simulations show that the plant and grid power loss models can be utilized in an optimal controller setting to increase the accumulated average efficiency (AAE). However, optimal controllers had slow prediction times, and therefore a Reinforcement Learning (RL) method, A2C, has been trained to learn an optimal control policy that maximizes system efficiency. The RL agent supersedes the optimal control techniques with up to 40 times faster prediction times.