Posters

Abstract

The large-scale integration of Renewable Energy Sources (RESs), namely wind and solar energy, into power systems is a major lever for the energy transition and decarbonization. However, their volatile production leads to a significant increase of uncertainty in power system operations. A Virtual Power Plant (VPP) aggregates RESs to operate them as a whole. The aggregation permits to smooth out variability and prediction errors as well as enhance the participation in electricity markets and the provision of ancillary services. A major question arises, how to efficiently schedule the available resources to maximize the VPP profit under the local constraints of each asset. Although there is a broad literature on these matters, the evolution of VPPs into large-scale aggregations integrating several types of Distributed Energy Resources (DERs), like storage, electrical vehicles, residential/industrial flexibility, hydrogen and others, brings new challenges. This study explores various control architectures, built respectively on centralized, decentralized and hierarchical optimization. First, we formulate the optimal scheduling of a generic large-scale multi-technology VPP on the day-ahead electricity market as centralized and decentralized mixed-integer programming problems. Second, due to the issues that a centralized architecture may create in terms of privacy, resilience and scalability, we explore three augmented Lagrangian relaxation (ALR)-based hierarchical control approaches: Alternating Direction Method of Multipliers (ADMM), Auxiliary Problem Principle (APP) and Analytical Target Cascading (ATC). These control strategies are compared using data from a real-world case study provided by the French aggregator Compagnie Nationale du Rhône. Results show the economic advantage of aggregating DERs, instead of operating them independently. Moreover, we show the benefits that hierarchical control provides in terms of scalability, protection of DER users' privacy and robustness to communication failures. Finally, hierarchical approaches are compared in terms of convergence properties and complexity of their implementation (e.g., parameters tuning) in the industrial environment.