Michaela Killian

Hierachical Fuzzy MPC Concept for Building Heating Control

Abstract This paper presents a hierachical model predictive control (MPC) structure with decoupled MPCs for building heating control using weather forcasts and occupancy information. The two level control structure embeds a fuzzy MPC (FMPC) for user comfort optimization and a mixed-integer MPC (MI-MPC) for energy optimization at minimal costs. As FMPC uses a set of local linear models classical linear MPC theory is applicable, though the underlying system dynamics is non-linear. The supply level in a large modern office building always features switching states of aggregates, hence an MI-MPC is used for energy supply optimization. Additionally, both FMPC and MI-MPC consider all relevant constraints. The innovation in this study, beside the usage of FMPC for building control, is the decoupling of the energy supply level and the user comfort with a single coupling node. Although a global optimum is not guaranteed, a decoupled control system often is more attractive for industrial applications and building operators. The perfomance of the proposed control structure is demonstrated in a simulation with a validated building model, and two different disturbance scenarios are presented.

Cooperative Fuzzy Model-Predictive Control

In this paper, a cooperative fuzzy model-predictive control (CFMPC) is presented. The overall nonlinear plant is assumed to consist of several parallel input-coupled Takagi-Sugeno (T-S) fuzzy models. Each such T-S fuzzy subsystem is represented in the form of a local linear model network (LLMN). The control of each local linear model in each LLMN is realized by model-predictive control (MPC). For each LLMN, the outputs of the associated MPCs are blended by the fuzzy membership functions, which leads to a fuzzy model-predictive controller (FMPC). The resulting structure is one FMPC for each LLMN subsystem. Overall, a parallel combination of FMPCs results, which mutually affects all LLMN subsystems by their respective manipulated variables. To compensate detrimental cross-couplings in this setup, a cooperation between the FMPCs is introduced. For this cooperation, convergence is proven, and for the closed-loop system, a stability proof is given. It is demonstrated in a simulation example that the proposed input-constraint CFMPC algorithm achieves convergence of the fuzzy LLMNs within few cooperative iteration steps. Simulations are given to demonstrate the effectiveness of the theoretical results.

Cooperative and hierarchical fuzzy MPC for building heating control

A combined cooperative and hierarchical control structure utilizing Fuzzy Model Predictive Control (FMPC) for building heating is introduced. The structure comprises three types of Model Predictive Controllers (MPC): For different independent zones of the building FMPCs optimize the fast acting input variable fan coils (FC) while a global linear MPC optimizes the slowly acting thermally activated building systems (TABS). Cooperation between these two groups of controllers is guaranteed by an inter-sample iteration. This cooperative structure acts as master in a hierarchical structure, where the slave is a mixed-inter MPC (MI-MPC) in the supply level. While the cooperative structure ensures user comfort in the building, the MI-MPC optimizes monetary costs of heat supply. This structure allows for decoupled and independent modeling of FMPCs, simple incorporation of the coupling input TABS, and decoupled design of the supply level control. A discussion on stability and sub-optimality of the control structure is given. A simulation of a large office building incorporating disturbances of ambient temperature, radiance, and occupancy demonstrates the performance of the proposed concept.

Pre-processing of Partition Data for Enhancement of LOLIMOT

The Local Linear Model Tree (LOLIMOT) algorithm is a versatile tool for black-box identification of nonlinear complex systems with a set of local linear models. In this work two methods for pre-processing of the partition data for this algorithm are presented. These methods aim at reducing the number of LLMs while improving the global model fit. The proposed methods are a (linear or nonlinear) principal component analysis and a rotational transformation of the input space. Both methods aim at mitigating the limitations of the axis-orthogonal splits in the partition space that LOLIMOT performs. The application to real data from industrial processes and the efectiveness is demonstrated on a grate-fired biomass plant and the thermal model of a large office building.

Cooperative Fuzzy model predictive control for a multivariate process

In this paper a cooperative Fuzzy model predictive controller (CFMPC) for a multivariate process is presented. Instead of designing one multivariate Fuzzy model predictive controller (FMPC) several more simple FMPCs for the input-coupled subsystems are designed, and optimal behavior is guaranteed by a cooperative iteration-loop. This ensures added flexibility in changing and extending the control structure since individual FMPC designs do not affect the other control loops. Moreover, the concept is ideally suited for combination with fault diagnosis and isolation (FDI) as no reconfiguration of the control structure is necessary. The positive features are demonstrated by a simulation of an input-coupled heating process, where over-actuation, actuator fault compensation and under-actuation are handled optimally by the CFMPC.

Verteilte kooperative modellprädiktive Temperaturregelung für komplexe Gebäude

ZusammenfassungIn dieser Publikation wird eine verteilt kooperative modellprädiktive Temperaturregelung für komplexe Gebäude präsentiert. Das nichtlineare Gebäudemodell wird mittels Takagi-Sugeno-(TS-) Fuzzy-Modellen mathematisch formuliert. TS-Fuzzy-Modelle sind lokal lineare Modellnetzwerke, welche als Gesamtheit ein nichtlineares Modell wiedergeben. Modellprädiktive Regelung von TS-Fuzzy-Modellen führt zu einem nichtlinearen Regelungskonzept, dem sogenannten fuzzy-modellprädiktiven Regler (FMPC). Wegen der speziellen Zonen-Modellierung von komplexen Gebäuden und der Abhängigkeit von Kopplungszonen, welche sich über ein gesamtes Stockwerk erstrecken, wird ein relaxierter kooperativer FMPC eingesetzt. Die Relaxierung zeichnet sich dadurch aus, dass die Kopplungszone getrennt von den Zonen-FMPC geregelt wird und als Störgröße in die Kooperation der FMPCs miteingeht. Neben der einfachen Trennung von Zonen ist dieses Konzept wegen der MPC-Struktur für komplexe Gebäude sehr geeignet, da dieses mit Beschränkungen in Ein- und Ausgangsgrößen optimal umgehen kann. Neben der Modellierung und dem Regelungsschema wird ebenso ein Simulationsbeispiel für ein spezielles Demonstrationsgebäude angegeben, welches die Vorteile dieses Konzepts hervorhebt.AbstractIn this work a distributed Cooperative Fuzzy Model Predictive Control (CFMPC) scheme for building heating control is presented. The dynamical non-linear building model is described by a local linear model network (LLMN). These LLMNs are described by Takagi-Sugeno-(TS-) Fuzzy-models. These TS-Fuzzy-models represent the non-linear complex building model with local linear models, which are easy to handle by a model predictive control (MPC) strategy. Because of the non-linearity the resulting MPC is given by a fuzzy MPC (FMPC). In this work a specific building is given, which is split into two zones and one coupling zone. The two zones are controlled by fan coils and the coupling zone is controlled by a thermally activated building system (TABS) which manipulates the temperature of the concrete core. Because of the coupling zone a relaxation and a cooperative control strategy were chosen. The cooperative iteration update defines a manipulated variable TABS as a disturbance for the other FMPCs and vice versa for the coupling MPC. Furthermore, MPC in general is effective for handling input and output constraints. Therefore, it is a useful tool for building control. Furthermore, a specific demonstration building is described and simulation results show the benefits of the presented controller strategy.