PhD student Paul Westernmann successfully defended his thesis today with only editorial changes. Thanks Dr. David Bristow, Dr. Bryony DuPont and Dr. Nishant Mehta for examining. Check out Paul's impressive publication list here: Google Scholar
The Energy and Cities group has succefully obtained a grant from PICS with the City of Victoria to deliver effective climate mitigation and adaptation solutions for municipalities. The project has started April 2020 and will run for three years.
Using simulation to understand the behaviour of buildings and energy systems.
Future holistic, integrated energy systems solutions span from buildings (which are now active players in energy markets) to district and city-level designs and national scale infrastructure. Simulation is the sole means of exploring the performance of new designs and concepts in a rapidly changing techno-economic context; all buildings and systems are unique (there are no prototypes) and simulation can explore the influence of future contextual parameters (energy prices, warmer climates, upgraded technologies) on new designs.
Exploring design spaces by examining trade-offs between multiple objectives.
The design space of possible systems and their performance is vast and intricate, making it impossible to discover the best solutions by trial-and-error or by exhaustive evaluation of all options. Computational design optimization (e.g. multi-objective genetic algorithms and mixed-integer linear programming formulations) are powerful tools for finding high-performing designs and exploring their sensitivity, robustness and resilience.
Leveraging developments in machine learning to better explore systems and data.
The next steps in achieving real progress in computational design aids will encompass statistical emulation of complex models and hyper-heuristic optimisers that learn how to solve problems better. Statistical emulators (also called meta-modelling) use methods like neural networks to approximate detailed models that are too time-consuming to run directly. Hyper-heuristics involves “optimising the optimiser”, either in advance using training data or during an optimisation.
Assistant Professor, Group Leader
G Baasch, P Westermann, R Evins Identifying whole-building heat loss coefficient from heterogeneous sensor data: An empirical survey of gray and black box approaches (2021) Energy and Buildings 241, 110889
This paper benchmarks seven different methods for characterizing building heat loss by estimating the whole-building heat loss coefficient. It uses a large set of synthetic data generated using EnergyPlus simulations. The methods compared include gray box approaches (e.g. estimating parameters of RC networks) and novel black box approaches (e.g. neural networks).
A Rahimzadeh, TV Christiaanse, R Evins Optimal storage systems for residential energy systems in British Columbia (2021) Sustainable Energy Technologies and Assessments 45, 101108
We explore the changes in optimal storage system sizing for on-grid, off-grid and 100% renewable scenarios. In on-grid systems, the results show that even the cheapest energy storage system is not feasible with the current cost of grid electricity. For off-grid and 100% RE scenarios, we look at the performance and cost that storage would have to reach to play a role in such systems.
P Westermann, TV Christiaanse, W Beckett, P Kovacs, R Evins besos: Building and Energy Simulation, Optimization and Surrogate Modelling (2021) Journal of Open Source Software 6 (60), 2677
This paper describes the besos Python library (LINK) along with the associated BESOS platform (LINK), which help researchers and practitioners explore energy use in buildings more effectively. This is achieved by providing an easy way of integrating many disparate aspects of building modelling, district modelling, optimization and machine learning into one common library.
P Westermann, R Evins Using Bayesian deep learning approaches for uncertainty-aware building energy surrogate models (2021) Energy and AI 3, 100039
In this paper, we train two types of surrogate models (dropout neural networks and stochastic variational Gaussian Process models) that provide estimates of model error alongside predictions of specific outputs. The surrogate model processes 35 building design parameters (inputs) to estimate 12 annual building energy performance metrics (outputs). We benchmark both approaches and show their accuracy to be competitive. Overall model errors can be reduced by up to 30% if 10% of samples with the highest uncertainty are replaced with values from the underlying high-fidelity model.
Westermann, P., Welzel, M., Evins, R. Using a deep temporal convolutional network as a building energy surrogate model that spans multiple climate zones (2020). Applied Energy (278) 115563.
In this study we use convolutional neural networks that process annual hourly weather files that consist of more than 150'000 values to predict building energy performance at any location in Canada in less than a second. This is a fast alternative to whole building energy simulation which provides performance estimates in minutes.
Bowley, W., Evins, R. Assessing energy and emissions savings for space conditioning, materials and transportation for a high-density mixed-use building. (2020). Journal of Building Engineering (31) 101386.
This paper explores the combined impact of building-related and transportation emissions in the context of avoiding urban sprawl. The proposed 'mothership' combines all the functions of a typical suburb into one super-high-density development, reducing both sources of emissions.
Westermann, P., Deb, C., Schlueter, A., Evins, R. (2020). Unsupervised learning of energy signatures to identify the heating system and building type using smart meter data. Applied Energy (264) 114715.
The study provides a method to automatically extract heating system and building type information from smart meter time series data only. The method relies on the concept of energy signatures.Westermann, P., Evins, R. Surrogate modelling for sustainable building design – A review. (2019). Energy and Buildings (198) 170-186.
This is a literature review summarising the most important works in the field of building energy surrogate models.
Forde, J., Hopfe, C., McLeod, R., Evins, R. (2019). Temporal optimization of affordable Passivhaus dwellings: is peak load a more reliable metric for cost optimal design? Applied Energy (261) 114383.
This paper applies a multi-objective optimisation approach to affordable Passivhaus delivery. One key finding is that reduced south facing glazing improves future resilience to overheating. Overalll, we demonstrate how multi-objective optimization facilitates evidence-based decision making.
A comparison of building energy optimization problems and mathematical test functions using static fitness landscape analysis. C. Waibel, G. Mavromatidis, R. Evins, J. Carmeliet, Journal of Building Performance Simulation, 2019.
Fitness Landscape Analysis is a very important but generally overlooked part of optimization and design space exploration. In this paper we explore the similarities between actual building energy optimization problems from the literature and the COCO testbed of mathematical test functions used in the field of computational optimization. We identify a large degree of variability between the landscapes of different building problems, but we are able to link clusters of problems to different test functions.
Surrogate modelling for sustainable building design - A review. P. Westermann, R. Evins, Energy and Buildings, 2019.
Surrogate models emulate an expensive high-fidelity building simulation model using statistical methods from machine learning. Once validated to approximate the detailed model well enough, a surrogate can be used to almost instantly predict the performance of a wide range of designs. This paper focuses on surrogates that predict aggregated design metrics (e.g. annual energy use) rather than time series data. We review 57 applications of surrogate modelling to sustainalbe building design, and analyse the modelling methods used, variables and outputs examined and the performance achieved. This information is summarized to serve as practical guide, making surrogate modelling accessible for future researchers.
Clustering and ranking based methods for selecting tuned search heuristic parameters. C. Waibel, R. Evins, J. Carmeliet. IEEE Congress on Evolutionary Computation, Wellington, New Zealand, 2019.
An architect goes amongst the computer scientists to talk about new methods for tuning the parameters of search heuristics! This paper presents two alternative methods for selecting tuned hyper-parameters of search heuristics for computationally expensive simulation-based optimization problems.
Co-simulation and optimization of building geometry and multi-energy systems: Interdependencies in energy supply, energy demand and solar potentials. C. Waibel, R. Evins, J. Carmeliet, Applied Energy, 2019.
This paper presents a co-simulation framework for the simultaneous optimization of building geometries and multi-energy systems using the energy hub approach, showing that this has significant benefits over optimizing each independently. The study shows that coupling multiple simulators into a common optimization and design workflow 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, essential interdependencies between the energy supply and demand side can be captured in the design of energy efficient cities.
Building Energy Optimization: An Extensive Benchmark of Global Search Algorithms. C. Waibel, T. Wortmann, R. Evins, J. Carmeliet, Energy and Buildings, 2019.
We present an extensive benchmark of black-box optimization algorithms for a range of building energy simulation problems, the aim being to go beyond 'yet another algorithm' that is demonstrated for only one problem. Using established and newly proposed metrics for optimizer performance evaluation, we show that no single optimizer dominates benchmark for all metrics and/or problems. There is some impact of tuning hyper-parameters of optimizers to specific problem dimensions. Overall, RBFOpt is the best algorithm for a very small evaluation budget, but best results for high evaluation budgets are obtained with CMA-ES.
Decarbonizing the electricity grid: the impact on urban energy systems, distribution grids and district heating potential., B. Morvaj*, R. Evins & J. Carmeliet, Applied Energy, vol. 191, pp. 125–140,
This paper determines the impact on district energy systems (DES) of different levels of renewable energy share in the electricity grid using a model that integrates DES optimisation, linearised AC power flow and district heating design.
Optimization framework for distributed energy systems with integrated electrical grid constraints., B. Morvaj, R. Evins & J. Carmeliet, Applied Energy, vol. 171, pp. 296-313, 2016.
Broad energy systems research regards the electrical grid as an infinite resource when optimising building and district-level systems, whereas power systems research conducts detailed analysis of the electrical grid, but treats the system components, layout and operation as fixed. This work brings together these two domains by incorporating (linearised) power flow equations into the energy hub modelling framework, optimising the energy system holistically whilst accounting for the limitations of the electrical grid. This gave significantly different results, highlighting the importance of accounting for these constraints.
Variability between domestic buildings: the impact on energy use, R. Evins, K. Orehounig & V. Dorer, Journal of Building Performance Simulation, vol. 9(2), pp. 162-175, 2016.
There are many sources of variation between different buildings, including occupant presence and behaviour, lighting and appliance use and temperature preferences as well as physical properties like insulation and air tightness. These are of particular importance when analysing district energy systems, since the relative temporal alignment of the loads has a huge impact on the aggregated profiles. A statistical analysis was performed on these profiles to identify the greatest sources of variation, giving a better understanding of how district systems should account for this at the design stage.
Multi-level optimization of building design, energy system sizing and operation, R. Evins, Energy, vol. 90 II, pp. 1775-1789, 2015.
In order to explore the performance of different energy system designs, it is necessary to know how they will perform operationally; the more complex the system, the more detailed this operational simulation must be to capture the true behaviour. This work nests both a building energy simulation (EnergyPlus) and an operational scheduling problem (the energy hub model) within a wider design exploration (a multi-objective genetic algorithm) so that the three levels can be explored holistically, leveraging the benefits of different solvers at each level.
New formulations of the energy hub model to address operational constraints, R. Evins, K. Orehounig, V. Dorer & J. Carmeliet, Energy, vol. 73, pp. 387-398, 2014.
The energy hub concept that sits at the core of much of my recent work balances intermittent energy demands and supplies through an operational optimisation every timestep. It can be applied from building to city scale, and can inform the design of many types of multi-energy systems and networks. However, in the simplified forms used prior to this work, the model fails to account for many significant aspects of equipment performance like minimum loads, run times or startup frequencies. This work adds rigour by including many additional constraints regarding the way energy systems really function, making this popular method much more applicable to the analysis of real energy systems.
A review of computational optimisation methods applied to sustainable building design, R. Evins, Renewable and Sustainable Energy Reviews, vol. 22, pp.230-245, 2013.
This review summarises the current state of the field of building optimisation, analyses the trends in the rapid growth of the field, and critically addressed the gaps in the existing body of work. I commented on possible future directions that might be profitably explored, and gives opinion on the best ways that this could be achieved. This work has become very highly-cited as an overview of the progress and challenges in the field.
A case study exploring regulated energy use in domestic buildings using design-of-experiments and multi-objective optimisation., R. Evins, P. Pointer, R. Vaidyanathan & S. Burgess, Building and Environment, vol. 54, pp. 126–136, 2012.
This paper applied the broad findings of my doctoral research to address the optimisation of domestic buildings in the UK to meet the building regulations as cheaply as possible – a situation faced frequently by industry practitioners. It uses design-of-experiments and systems concepts to reduce the scale of the problem to make it computationally manageable, also critical for application in real practice. The underlying simulation process was encapsulated in a software tool that could be easily applied by design team members.