[Back]

M. Killian, M. Kozek, B. Mayer, M. Mayer:
"Eine flexible Methodik für die modellbasierte prädiktive Regelung komplexer Gebäude";
Talk: enova 2015 Internationaler Knogress: Nachhaltige Gebäude, Pinkafeld (invited); 2015-11-26 - 2015-11-27; in: "Nachhaltige Gebäude", FH Burgenland GmbH, 19 (2015), ISBN: 978-3-7011-0350-8; 405 - 420.

Model based control of complex buildings requires several obstacles to be overcome: A suitable model for control design must be of comparatively low order and preferably of linear structure. The model must comprise all manipulated variables and all disturbance inputs. Many disturbance inputs are stochastic, like ambient temperature, irradiation, and occupancy. Building zones with a dedicated manipulated variable or with generally differing disturbances (like north or south fronts) need to be modelled by a dedicated zone. Also strong couplings as established by thermally activated building systems need to be considered separately. Additionally, analytical modeling typically leads to a high order nonlinear model with several hundreds of unknown parameters which would have to be fitted to measurement data. Moreover, modeling of the energy supply level leads to hybrid system equations as switching aggregates like refrigeration machines and heat pumps are operated alongside continuous supplies such as district heating and alternative heat sources like geothermal supply.
In this paper a staged approach is proposed: The overall building is split into two entities - the user or comfort area and the energy supply systems. The user area is split into several zones according to the differentiation given above. The models for these zones are based on measured data, hence black-box models result. In the case of incorporating expert knowledge, the resulting models are called gray-box models. These models are either linear in structure (especially if only used for either heating or cooling) or they may also be nonlinear. A suitable globally nonlinear model structure for control design is given by local linear model networks, which enable the use of linear control design methods and the associated advantages like guaranteed stability. The energy supply level is mostly modelled by analytical relations, only some complex parts are incorporated by black-box models.
The proposed flexible control design is also based on this staged approach: A hierarchic model predictive control (MPC) design is proposed, where the MPC for the user area computes the minimal energy supply necessary for best user comfort, while the MPC for the energy supply guarantees the provision of the supply energy while guaranteeing minimal costs. This separation of goals achieves sub-optimal global performance, but the two parts of the hierarchical control may be implemented independently, only in the user area or only in the energy supply level. In these cases the single implemented part can cooperate with the ex-isting conventional structure. This adds to the flexibility of the implementation. The MPC for the user area is designed as cooperative fuzzy MPC while the controller structure for the en-ergy supply is chosen as a hybrid MPC with mixed-integer optimization. Some simulation re-sults for this hierarchical-cooperative MPC structure applied to a modern office building are given, demonstrating the effectiveness and high energy saving potential of the proposed control design. Some implementation aspects are also covered, as the control structure has been implemented in the process control system XAMControl by evon automation GmbH.