Model Mismatch Mitigation in Controlling District Heating Networks : Exploration of methods to mitigate the effects in control and feedback of model mismatch in the context of district heating networks with low computing and memory resources

Sammanfattning

District heating networks help reduce greenhouse gas emissions by integrating renewable energies like solar thermal energy. However, the integration of these intermittent energies requires advanced control, like Model Predictive Control (MPC). This type of control requires significant computational and memory resources, which are often limited. To alleviate those constraints, in this work, predictive control is transformed into an Energy Management System (EMS). Control variables considered by the EMS are typically at the thermal power or energy level rather than at the thermal-hydraulic level (temperature and flow rate). Temperature and flow-rate are the natural operational variables for optimizing production. Consequently, an intermediate controller is needed to correct modeling errors and mitigate the effects in control and feedback of reducing the complexity of predictive control. The effects in control come from the fact that the EMS provides average power setpoints. Likewise, feedback must be expressed in terms of power or energy (in general it is expressed in terms of temperature and flow rate). Therefore, it is necessary to estimate the energy stored in, for example, the thermal storage. Two methods to track average power setpoints have been tested through simulation and experimentation: the instantaneous correction method and the horizon correction method. Tests on a heat pump and a gas boiler conducted in this work showed that the horizon correction method is more accurate i.e. has a lower absolute error. This seems to be due to its ability to slightly adjust the instantaneous power setpoint to achieve the average power setpoint. For thermal storage energy estimation, three empirical methods have been compared: the hyperbolic tangent method, the linear method, and the layer method. Although the hyperbolic tangent method is the most accurate in simulation (i.e. has a lower relative error), the linear method was the most accurate experimentally. In conclusion, this work has demonstrated that, for the considered physical system, an intermediate control approach, combining the horizon correction method and a linear approximation of stored energy, mitigates satisfactorily the effects of the simplified model used by predictive control. This work paves the way for further research, such as optimizing the control of the distribution network with limited computational and memory resources or compensating for prediction errors in Model Predictive Control to consistently meet heat demand.