Simulation-based optimization of buildings and multi-energy systems

Environmental challenges such as climate change and resource scarcity call for more sustainable anthropogenic activities. Especially in the built environment it is of utmost importance to make the best possible decisions as early as possible due to the longevity and high environmental, social and economic impact of buildings. One crucial aspect for increasing urban energy efficiency is the promotion of multi-energy systems, as this allows synergetic effects in the generation, storage and conversion of multiple energy carriers. Due to the inherent complexity of buildings and energy systems, the decision-making process should be assisted by the application of simulation programs and computational optimization algorithms. As a consequence, this thesis revolves around research questions related to a simulation-based and optimization-driven design process of buildings and multi-energy systems, covering optimization algorithms and simulation models and their application in co-simulation design case studies. The algorithmic focal point of this thesis lies in search heuristics (also known as metaheuristics, derivative-free optimization, or black-box optimization). Such algorithms have shown to be of high practical benefit in engineering disciplines, as they do not require any information about the cost function, other than the cost value itself when queried with a decision variable vector. This is useful when complex simulation programs and geometry variables are included, such as in building energy optimization (BEO). However, there is a lack of knowledge on (i) how to properly set algorithmic parameters (hyper-parameters), (ii) when to use which of the many available search heuristics, and (iii) which properties of BEO problems may be exploited for answering (i) and (ii). These three points are investigated in this thesis by developing approaches to (i) “optimize the optimizer" (i.e. tuning hyper-parameters) to improve algorithm performance, (ii) by benchmarking different optimizers on a building energy test bed to identify the most suitable optimizer in a building engineering context, and (iii) by analyzing and comparing problem structures of typical BEO problems to learn about recurring patterns that may be exploited in future algorithm development and their real world application. Furthermore, as a foundation for (i) this thesis develops a consistent algorithm framework for search heuristics that allows comparing and hybridizing algorithmic operators, thus allowing an in-depth understanding of how search heuristics are designed and may be developed further to better solve BEO problems. The findings show that the hyper-parameter tuning methods developed are successful as they resulted in search heuristics with competitive performance to other popular and/or modern algorithms such as CMA-ES and RBFOpt. Generally, we show that classical randomized algorithms (such as Evolutionary Algorithms and Particle Swarm Optimization) are especially good with large evaluation budgets and at very difficult BEO problems, and surrogate model-based solvers (RBFOpt) are especially beneficial with very low evaluation budgets and less difficult BEO problems. The second focus of this thesis is on the development of simulation models and combining and implementing them in a co-simulation and optimization environment. This includes the development and validation of an efficient three-dimensional solar potentials model and of an airflow model for early design stage. The formulated co-simulation optimization problem combines the optimization of building geometry, the assessment of solar potentials on roofs and fa cades, the simulation of building energy demands, and the selection and sizing of an optimal multi-energy system. This allows the study of interdependencies between the demand and supply side. The knowledge and experience from the previous investigations on search heuristics are used for solving a building design problem of four office buildings in the city of Zurich. Results of the co-simulation case study show the importance of considering multiple design aspects simultaneously, as only in this way can their interdependencies be identified and exploited. We predict that coupling multiple simulators into a common optimization and design work flow will become more common as it brings together architectural aspects (such as geometry) with engineering aspects (such as the energy system design) and microclimate conditions (such as local solar potentials), thus capturing essential interdependencies of real world problems. Search heuristics are indispensable for solving such complex problems. This thesis provided answers from the algorithmic and application perspectives that will enable the design of more energy efficient buildings and cities.