Talk @ FUSED 2023
- conferences
Johannes van Randenborgh presented his talk “A tailored model for sustainable control of ATES systems with mixed-integer programming” at FUSED 2023 in Bochum, Germany.
The talk by Johannes was presented at the conference “FUSED: From Underground to Surface: Energy provision and Distribution” organized by the Research Department of Subsurface Modeling and Engineering (RDSME) from Ruhr University Bochum. We would like to thank the organizing committee for having the opportunity to present our latest research at FUSED 2023.
Aquifer thermal energy storage (ATES) systems are used to temporarily store heat or cold in open aquifers in order to regulate building temperatures. Although the basic concept of storing heat in summer for winter (and vice versa) is simple, the efficient operation of ATES is non-trivial. For instance, ATES are necessarily combined with conventional heating systems or heat pumps to handle peak loads, which complicates their efficient operation. Moreover, it is important to maintain a certain heat balance in the underground to ensure long-term operation of ATES. Both challenges can be addressed with modern control technologies. In particular, model predictive control (MPC) enables to optimize the current operation while taking constraints and long-term requirements into account.
The performance of MPC crucially depends on the quality of the model. In fact, the model should accurately capture the dominant system dynamics while being numerically cheap to evaluate. Existing approaches often address only one of these aspects. For instance, Rostampour et al. (2016) consider a simplistic battery-like model whereas Beernink et al. (2022) build on a complex MODFLOW model. In our contribution, we present a novel MPC scheme which reflects a sweet-spot between these extremes. More precisely, our model builds on linearizations of the heat transport equation for the three operation modes injection, extraction, and storing (or inactivity). Incorporating these modes in the MPC leads to a mixed-integer (optimization) program, which can be solved efficiently compared to an MPC utilizing a MODFLOW model. This is illustrated with a numerical case study showing the effectiveness of our approach.