Benchmarking optimization problem formulations for Model Predictive Control of District Heating systems with a Software-in-the-Loop approach

The fourth generation of district heating (4GDH) is characterized by various generation technologies, particularly renewable energy and waste heat, and heat storage to use renewable heat when available, regardless of demand. Hence, efficient and predictive operational planning is becoming important. Model predictive control (MPC) is commonly proposed for this purpose. When modeling the optimization problems within the MPC, the individual components are often linearized. This enables a fast and unambiguous solution at the expense of accuracy. By contrast, the use of non-linear formulations can lead to a significant increase in complexity and thus reduce computational tractability, which impairs real-time capability. Given this trade-off between accuracy and computational challenges, there is a need to benchmark the performance of different model formulations against each other and against traditional control strategies.
In this work, a benchmarking approach using a complex transient thermo-hydraulic system simulation model (CTM) in real-time software-in-the-loop (SIL) operation is demonstrated. The flexible heating grid of the research facility "District LAB" at Fraunhofer IEE serves as the demonstration object. First, the CTM is implemented in the simulation environment Simscape, including a control structure to represent the reference case of rule-based heat storage charging control. Then, two different optimization problems are formulated using the Pyomo library in Python: a mixed-integer linear programming (MILP) formulation based on energy balances and a mixed-integer nonlinear programming (MINLP) formulation based on mass balances. Subsequently, a bidirectional data exchange between simulation and optimization is established.
Benchmarking is performed in SIL operation using input parameters from different seasons. The performance scores are the main outcome of this work, and it is expected that both optimization methods outperform the reference case. Furthermore, a quantification of the performance benefits is possible, which can serve as decision support for grid operators in the transformation process towards 4GDH. To further substantiate the results, additional experimental investigations at the District LAB facility are planned beyond this work.