Planning and Operation of Isolated Hybrid Power Systems

The ongoing energy transition motivates the deployment of ever-growing rates of renewable energy in global power grids. However, the inherent stochasticity of the solar and wind resources brings well-known challenges to the planning and operation of power systems. Furthermore, the decarbonization attempts of other energy sectors results in an electrification increase, which requires massive installation and causes transmission congestions. Thus, justifying the important role of energy storage systems and sector coupling as flexibility providers.
Hybrid power systems consist on a mix of traditional and renewable generators, energy storage, and demand responsive units. Clearly, they require complex control strategies to coordinate the mix of different technologies. In this context, islands represent excellent study cases among other reasons due to their isolation, great renewable resource, and limited size. Particularly, in this thesis, the archipelago of Cape Verde is used as testing ground for all developments due to their goal of reaching 100% renewable integration. A complete open access dataset including information on its whole energy system (electricity, transport, etc.); is presented to this very end.
Regarding planning, an optimization formulation enabling the design of future integrated energy systems with focus on flexibility is proposed. Its computational efficiency allows to run significantly longer horizons that those available in the literature with high resolution. This mixed-integer linear programming problem allows to configure and size any power system to satisfy the generation and storage requirements. Regarding sector coupling, it allows to either size, or only employ demand response as a flexibility source.
Then, operation requires an energy management system capable of handling uncertainty at different horizons and resolutions. A stochastic unit commitment formulation is proposed in this thesis focusing on flexibility exploitation under uncertainty in order to coordinate all different unit types. The necessary forecasts are developed as data-driven tools based on physics-informed machine learning. While the stochastic scenario configuration is approached as an unsupervised error characterization process of the forecasters. Then, a control architecture suitable for implementing complex energy management systems in hybrid power plants and systems is proposed. This is complemented with a scheduling tool coordinating conflicting results for different horizon-resolution combinations and a hierarchical structure meant to deal with unfeasible states.
Overall, flexibility is a key resource dealing with uncertainty, which in the planning stage comes from prices, demand evolution, and electrification; thus driving required energy storage capacity. During operation, stochastic optimization enables a most resilient implementation of generation, storage and sector coupling coordination. The uncertainty-dominated scenarios can be configured using machine learning-based fore- casters combined with unsupervised error characterization. Lastly, the implementation requires a suitable control architecture coordinating the energy management of technologies, horizons and resolutions.