Data-driven Renewable Energy and Storage Optimization in Integrated Energy Systems

Abstract

The growing integration of renewable energy sources into the power grid necessitates innovative approaches to system energy operation and scheduling. Integrated electricity and gas networks provide a promising solution to this challenge, enabling the efficient, reliable, and sustainable operation of energy systems. This thesis presents a novel data-driven framework for the optimal scheduling of integrated energy networks, addressing the challenges posed by high penetration of renewable energy sources.

A learning-assisted methodology is developed by integrating Graph Convolutional Network (GCN) and Bayesian-based uncertainty models to improve the accuracy and efficiency of scheduling integrated energy systems. The proposed GCN effectively captures complex interactions within the integrated energy networks, facilitating accurate predictions of nodal power and gas flows. Meanwhile, the Bayesian-based model adeptly manages the inherent uncertainties associated with renewable energy generation, employing a chance-constrained approach to ensure system reliability.

The effectiveness of the proposed methodology is validated through extensive simulations on an IEEE 39-bus electricity network coupled with a 22-node gas network. The results indicate significant improvements in predictive accuracy and computational efficiency compared to traditional model-based methods and existing data-driven techniques. This thesis contributes a hybrid learning-assisted framework that provides a foundation for intelligent and scalable energy management in integrated energy systems.