Gradient projection-based online feedback optimisation for distribution grid management

The increasing use of renewable energy sources within distribution grids has raised concerns about grid stability, power quality, and congestion. Especially, the share of energy generated by photovoltaics has grown considerably. However, the optimal use of their inverters has also been proposed as a possible solution to mitigate such concerns through reactive power compensation and curtailment. This paper introduces several optimisation algorithms for online feedback optimisation and states their advantages and disadvantages. A projected gradient descent algorithm is developed and tested under practical considerations such as com- munication delays, measurement errors, and modelling inaccuracies in the form of Gaussian noise. Simulations were performed on a medium voltage distribution grid model using exact AC power flow. The findings suggest that the algorithm is robust to measurement noise and modelling errors but it is susceptible to communication delays.