[en] A new body of research has recently emerged in energy systems, focusing on the links between centralized and decentralized power generation, and on the interactions between different energy sectors. This research has brought new concepts such as « Smart Energy Systems », « Integrated Energy Systems » or « Virtual Power plants ». In smart energy systems, the focus is on the integration of the electricity, heating, cooling, gas and transport sectors, and on the use of the flexibility in demands and various short-term and longer-term storage. To enable this, the smart energy system must coordinate between a number of sectors, which includes electricity grids, district heating and cooling grids, gas grids and different fuel infrastructures.
Modeling these different aspects in an integrated manner is a challenge: the high temporal and technical granularities required by the power system models can become uncompatible with the requirements of long-term energy planning involving all relevant energy sectors. Current solutions to this involve hard-linking or soft-linking between models, with their associated challenges such as computational traceability or numerical convergence.
In this paper, an alternative approach is proposed to couple a power system model (the EU-wide Dispa-SET power system model) and a system dynamics model (the MEDEAS model). The methods relies on the creation of surrogate models that approximate the results of the power system optimizations. A multi-dimensional inputs space is created by varying key system characteristics: flexible capacity, no flexible capacity, short-term storage, long-term storage, grid infrastructure, renewable penetration. A latin hypercube sampling is then defined to run the model over this inputs space and Artificial Neural Networks are used to predict key system performance indicators (in this case curtailment and loss of load) as a function of the system features.
Results indicate that ANN can predict with good accuracy the main power system constraints and outcomes. The generated surrogate models are therefore suitable to be integrated into a more general system-dynamics model of the energy systems, thus improving the representation of the power system operation and constraints.