Author:

Abstract:

A large potential exists to improve the current practice of HVAC design and operation of buildings with respect to occupant comfort, energy use or energy cost, and investment costs for design and construction. More specifically, design and control processes can be improved through the use of contemporary optimisation algorithms such as Model Predictive Control (MPC). Numerous papers have demonstrated the value of MPC using both simulation and demonstration projects. Practical implementation of MPC in industry is however hampered by the expert knowledge and time investment that is required for developing MPC controller models and algorithms. An MPC development methodology that is both practical and scalable to large problem sizes has not been demonstrated to date. Moreover, design studies typically do not take into account the interaction between control and design. This PhD thesis therefore proposes a user-friendly, object-oriented methodology for integrated optimal control and design of buildings. For this methodology, modelling experts develop generic, detailed, but easy to use component models for optimal control problems using the object-oriented modelling language Modelica. Users without expert knowledge combine these models into system diagrams through the use of simple connections and physically interpretable model parameters. Computer algorithms then infer the problem structure and equations from the connected components, and automatically generate efficient code for solving the equations and for minimizing a user-defined cost function, while also satisfying technical and comfort constraints. This methodology for optimal control can be applied to many designs automatically such that the optimal design, when using an optimal controller, is identified for a user-defined set of design variables. Whereas the use of detailed component models typically leads to large computation times, efficient algorithms and methodologies are proposed such that such convenient models can be used nonetheless. Following contributions to 1) simulation, 2) optimal control and 3) optimal design have been made to implement the outlined methodology: The IDEAS, Annex 60 and IBPSA Modelica libraries have been developed further such that their models are sufficiently detailed for building energy simulations. These libraries have also been modified such that they lead to more efficient and more robust simulations and new models have been added. Furthermore, generic guidelines are presented for the development and efficient use of these models. The accuracy and numerical efficiency of the models is demonstrated through the implementation of the simulation model of a 32-zones, well-insulated office building, Solarwind. Computation speed increases of about three orders of magnitude are obtained relative to a case where the guidelines are not applied. The IDEAS and IBPSA libraries have been modified or extended to allow optimisation using the same, or slightly modified, models: derivative-based optimisation algorithms require that the cost function and constraints are twice continuously differentiable with respect to the optimisation variables. A Toolchain for Automated Control and Optimisation (TACO) is presented based on the open-source software JModelica. TACO automatically translates the Modelica code for one specific building design into an MPC problem. The resulting problem is optimised using the open-source derivative-based optimisation algorithm IPOPT. The optimisation problem is formulated efficiently using CasADi, such that the computation speed is much higher than the state-of-the-art. TACO is applied to the case study model to demonstrate the usability and scalability of the approach. Moreover, for this case study, the MPC uses 82 % less electrical energy than a state-of-the-art rule based controller, at comparable thermal comfort levels. However, the thermal balance of the borefield is not maintained for this case study, meaning the energy savings may decrease after a few years. This can be resolved in future research. The achievable energy savings depend on the building and the reference that is compared with. An integrated optimal control and design methodology is presented that generates designs for which the operational and investment cost are computed using TACO. The methodology is applied to the Solarwind case, where the net present value of the operational cost and investment for the optimised design variables is reduced by 65 %.