Posted in: News
Consortium members, Filip Jorissen, Wim Boydens & Lieve Helsen have recently published a paper on TACO, an automated toolchain for model predictive control of building systems: implementation and verification, in the Journal of Building Performance Simulation
TACO will be further developed and demonstrated during the coming years with the goal to significantly reduce the engineering effort required for developing MPC for building applications.
Thermal systems in buildings account for approximately 15 % of the primary energy use. Model Predictive Control (MPC) is a promising approach to optimally control these thermal systems and thus to achieve energy savings. Research on MPC can be categorized based on the level of physical knowledge and measurement data that are used to create the controller. White-box models do not use measurement data to create a model and rely on physical knowledge of the controlled system whereas black-box models rely on measurement data and grey-box model combine the two. The bulk of the research focusses on grey-box and black-box control methods, usually for fairly small demonstration cases and requiring extensive post-processing or an existing rule-based controller. We however argue that, for any MPC approach to be economically viable in industry, there are a set of requirements. 1) The approach should be scalable to and demonstrated using large, complex buildings. 2) The approach should be able to start without measurement data since this data is not available for new buildings. 3) It should be systematic and easy to use without requiring model tuning by an experienced engineer. 4) It should have systematic performance without failure.
Given these set of requirements, using a typical grey-box or black-box approach is not obvious. Instead, we propose a white-box based approach using detailed Modelica models. These models are automatically compiled into an optimization code using the recently developed Toolchain for Automated Control and Optimization (TACO). White-box models have the tendency to require a lot more computation time, but this hurdle has been overcome through the efficient implementation of TACO.
Our presentation will present TACO and its advantages over other approaches for implementing MPC. Furthermore, results are presented from the first real demonstration of TACO on an office building in Belgium. Finally, we discuss further development potential of TACO and white-box based models in general, which includes optimal control, optimal design and fault detection of buildings, including system integration with thermal networks and distribution grids. White-box models can thus ultimately lead to a new way designing and constructing the built environment.
This can be read here
This paper presents TACO (Toolchain for Automated Control and Optimization), which is a Modelica-based automated toolchain for model predictive control (MPC) of building systems. Its goal is to significantly reduce the engineering expertise and the time investment required for applying MPC to buildings. TACO is based on JModelica. Modifications compared to JModelica are discussed and the implementation of our custom MPC problem formulation is presented. The implementation is verified using two example models and is benchmarked with respect to accuracy and computation time. These results show that the computation time can be reduced significantly using the toolchain options, while only slightly reducing the controller optimality.