Distributed Control Strategies for Distributed Storage

Consulta en la Biblioteca ETSI Industriales (Riunet)[EN] In the next few years, it can be expected that many battery systems will be installed in the distribution grid to cope with the high in-feed of photovoltaics. In particular, battery systems can help stabilize the power system by managing short-term power uctuations caused by sudden changes in demand and supplies from RES, as well as reduce grid's overall operation costs. However, the time-varying state of charge from batteries induces a time dependent behaviour which leads to a greater complexity of the optimization problem with hard energy capacity constraints and dynamic couplings. Therefore, new control strategies are needed to dispatch a large number of battery systems. In this Thesis, to reduce the problem complexity, a distributed control strategy based on the OCD algorithm is developed and applied to interconnected LV grids. This metholodogy divides the central dispatch problem into several smaller iteratively solvable subproblems, which are solved independently but coordinated whereas taking the grid constraints into account. Each subproblem comprises one LV grid and is coordinated with the neighbouring LV grids by exchanging updated variables and Lagrangian multipliers from the bordering nodes. This exchange is done after one iteration. In each LV grid, PV and battery systems are assigned to households and distributed over the grid nodes. Additionally, to deal with forecast uncertainties and batteries' dynamic behaviour, the proposed approach has been extended by applying a Model Predictive Control which allows the optimization problem to be solved over a prediction horizon and handle the dependency between consecutive time steps. As an outcome of this thesis, the developed distributed control strategy is compared with the centralized problem in terms of computation e ort and optimality. The simulation results show that the distributed control problem converges to the centralized solution, however, a higher number of iterations is needed. For large-scale systems, a parallel algorithm presents the advantage of needing less computation time than a centralized strategy. As the interior point algorithm is used, a high dependency is seen between convergence and the barrier term. Furthermore, due to the radial structure, convergence is strongly related to the feeder's control variables for time steps with low PV generation.Hafeli, P. (2015). Distributed Control Strategies for Distributed Storage. http://hdl.handle.net/10251/69737