Model Predictive Control of Thermally Active Building Systems and Monitoring of two Test-Boxes

Starting point/Motivation

Thermally Active Building Systems (TABS) can be utilized as short-term storage for heating or cooling. Increasing cooling demand for air conditioning leads to extended operation of air conditioning systems during the day. In general, high outside temperatures have an adverse effect on the air conditioning system efficiency, the economic costs and cause peak loads in the grid. In addition, district heat main supply pipes operating on their limit often prevent from further extension of district heating in remote areas. Reduction of peak load in buildings through temporal extension of cold and heat delivery throughout day and night time requires a predictive control concept making use of weather forecast data. Such a controller also eases the utilization of renewable energy. The enormous complexity of existing approaches for predictive building automation prevents this technology from being used in practice. Although it has been investigated in a number of research projects official numbers as to which level energy can be saved are not yet available. On the other hand, practitioners claim energy savings in a broad range owned to the utilization of predictive control, but the quality of the used weather forecast data in practice is very low and the predictive control approaches are rarely described in depth. A standardized method for thermal predictive control of TABS is missing as is an experimental set-up for investigation of various controllers for this purpose.

Contents and Objectives

The aim of this project is to build two "Test-Boxes" being placed in real ambient conditions, and successfully apply a model predictive controller for a TABS – being part of each Test-Box – in one Test-Box. The given constraints are: efficient energy utilization, maximization of thermal comfort and robustness. A predictive controller utilizing a simple model of the control process and weather forecast data as input data is planned and realized for this purpose. In addition, the complex model aids to optimally place sensors, which will be used to frequently assimilate the simple model during real time operation of the predictive controller. The second Test-Box operated with a standard hysteresis controller serves as reference case. To monitor actual values, the Test-Boxes and the TABS will be equipped with a number of sensors. Maximum accuracy for in situ weather data will be attained placing a weather station close to the Test-Boxes; obtained data will be used to investigate different approaches for in situ weather forecast optimization.

Methods

The research project is divided into four work packages:

1. State of the art, Problem Analysis, Solution Concepts

In this phase of the project, literature is screened and a concise literature review is compiled. Furthermore, results and findings from relevant preliminary projects are collated. Within the framework of a workshop, to which developers and users in the area of thermal building conditioning are invited, topics such as research needs are discussed and the exchange of experiences concerning the research topic takes place. In addition, the opinion of experts on different topics is brought into the planning-phase of the Test-Box set up. In the case of weather forecasting, an overview of the operation of the different methods for site optimization of weather forecast data is provided.

2. Test box assembly, system identification, weather forecast

Within the scope of this work package mechanical, electrical and hydraulic assembly work is necessary for the construction of the test boxes and the installation of the weather station. This work also requires appropriate commissioning for various trades. In addition, the planned measurements require the installation of IT hardware (database, set-up of various interfaces) and the statistical data evaluation.

3. Measurement, simulation, controller analysis

The methods in this work package are similar to those from the previous work packages. It is also necessary to compare the test results from the two test boxes for different controllers and additional modelling and simulation for controller design and optimization. Statistical data analysis plays a central role. Preparations for the last work package are similar to those in the first work package.

4. Integral testing and method evaluation

The methods are similar to those from the previous work package and relate to the evaluation of the results. To obtain statements about expected behavior and implications for similar applications, theoretical considerations and interpretations are necessary. Within the scope of this work package the economic costs are also touched.

Results

A major result is predictive energy efficient heating and cooling by means of a TABS without any losses of the level of comfort (undercooling). One aim is the reduction of energy demand – required for cooling – by at least 10%, for the predictive controlled Test-Box. The project should provide insight to parameterization rules of a predictive controller for TABS, dependent on easily available physical parameters. Another important result is the identification of the optimal sensor positions for the purpose of model assimilation, to allow for a simple thermal model as part of the predictive controller. A maximum of transferability is aimed at to assure usability for similar applications in different environment.

Prospects / Suggestions for future research

The end of this project marks and suggests the beginning of a new project. This is especially true because the results with respect to the potential energy savings are motivating, and second the findings on the subject of MPC with respect to buildings (building dynamics model, cost function) within the framework of the simple real test environment give confidence for a larger, more complex building project with high success potential. The content to be treated would be, above all, the question how to link prototypical thermal zone models (as designed and examined for the controller in this research) in the context of a more complex controller model for a large building, and special features of the model parameter identification in this case.

Concerning a practical implementation in commercial style, it still needs to be clarified whether tools and algorithms as used in the project (MATLAB, MPC-Toolbox, YALMIP, Gurobi-Solver) can at least be partially used in this way, or whether a partial or complete self-development of the necessary algorithms is to be preferred.