No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Applying Deep Learning and Databases for Energy- efficient Architectural Design Abstract
Abstract
manavmahan.singh@kuleuven.be 2,3,4 {patricia.schneider|hannes.harter|w. lang}@tum.de 5 geyer@tu-berlin.de The reduction of energy consumption of buildings requires consideration in early design phases. However, modelling and computation time required for dynamic energy simulations makes them inappropriate in the early phases. This paper presents a performance prediction approach for these phases that is embedded in a multi-level-of-development modelling approach. First, parametric pre-trained modular deep learning components are embedded in the building elements. The energy performance is predicted by composing these components. Second, embodied energy assessment is performed by extracting the information from a database. A calculation module queries the database and calculates the embodied energy. Both, embodied and operational, energy are assembled to predict lifecycle energy demand. The method has been implemented prototypically in a digital modelling environment Revit. A case study serves to demonstrate the application process, the user interaction and the information flows. It shows energy prediction in early design phases to enhance the environmental performance of the building.
... However, this approach has been applied only to predict deterministic energy demand using simulation tools such as EnergyPlus [38] and Modelica [39,40]. Probabilistic energy predictions have been rarely coupled with BIM tools that require a novel approach to process the uncertain design information and make predictions for several combinations [18,41]. The current approaches to obtain deterministic BPS results needs adaptation for probabilistic BPS results using ML models. ...
Global energy concerns necessitate designing energy-efficient buildings. Many important decisions affecting energy performance are made at early stages with little information. Dynamic simulations support informed decision-making; however, uncertainty, high computational time, and expensive modelling efforts impair their use at early stages. This article develops an approach using building information modelling and machine learning that provides quick energy performance information. This approach has been implemented into a web tool, p-energyanalysis.de. It allows design space exploration, assesses the energy performance of design options, compares multiple options, performs sensitivity analysis, and tracks changes. Twenty-one participants (researchers and architects) used it as a support tool for designing an energy-efficient building. Their feedbacks are discussed as part of the tool development. The study found that the tool supports early-stage design decisions by quickly providing relevant information. The limitations, such as the bias in the results towards training data population and implementation issues, are also discussed.
The early stages of building design involve the consideration of different design variants and their assessment regarding various performance criteria including energy consumption and costs. During the design process, the involved experts from different disciplines frequently exchange building information to develop a design that satisfies the project's requirements and objectives. In the course of this iterative process, the building design evolves throughout multiple refinement stages. At the same time, different variants are developed. In BIM-based projects, the maturity of the design information provided by the model is expressed by the notion of Level of development (LOD). So far, however, there is no method to formally define the information requirements of a LOD. In particular, there are no means for expressing the uncertainty involved with the provided information. By contrast, despite the insufficient information available in early design stages, a BIM model appears precise and certain. This situation leads to false assumptions and model evaluations, for example, in the case of energy efficiency calculations or structural analysis. Hence, this paper presents an overview of a set of approaches that were developed to alleviate and preserve the consistency of the designed solutions. The approach includes the development of a multi-LOD meta-model, which allows one to explicitly describe the LOD requirements of each building component type incorporating the possible uncertainties, e.g. concerning the building dimensions. On the basis of this multi-LOD model, methods for evaluating a building design's performance regarding the building's structure and life cycle energy performance are proposed that take the defined uncertainties into account. To support the management of design variants in one consistent model, a graph-based approach is introduced. Finally, a minimized communication protocol is described to facilitate the workflow and communicate the evaluation results for supporting the decision-making process.
With current efforts to increase energy efficiency and reduce greenhouse gas (GHG) emissions of buildings in the operational phase, the share of embedded energy (EE) and embedded GHG emissions is increasing. In early design stages, chances to influence these factors in a positive way are greatest, but very little and vague information about the future building is available. Therefore, this study introduces a building information modeling (BIM)-based method to analyze the contribution of the main functional parts of buildings to find embedded energy demand and GHG emission reduction potentials. At the same time, a sensitivity analysis shows the variance in results due to the uncertainties inherent in early design to avoid misleadingly precise results. The sensitivity analysis provides guidance to the design team as to where to strategically reduce uncertainties in order to increase precision of the overall results. A case study shows that the variability and sensitivity of the results differ between environmental indicators and construction types (wood or concrete). The case study contribution analysis reveals that the building's structure is the main contributor of roughly half of total GHG emissions if the main structural material is reinforced concrete. Exchanging reinforced concrete for a wood structure reduces total GHG emissions by 25%, with GHG emissions of the structure contributing 33% and windows 30%. Variability can be reduced systematically by first reducing vagueness in geometrical and technical specifications and subsequently in the amount of interior walls. The study shows how a simplified and fast BIM-based calculation provides valuable guidance in early design stages. Keywords: early building design; life cycle assessment (LCA); building information modeling (BIM); embedded greenhouse gas emissions; embedded global warming potential; life cycle energy analysis; life cycle energy assessment; design assessment; embedded primary energy
Increasing sustainability requirements make evaluating different design options for identifying energy-efficient design ever more important. These requirements demand simulation models that are not only accurate but also fast. Machine Learning (ML) enables effective mimicry of Building Performance Simulation (BPS) while generating results much faster than BPS. Component-Based Machine Learning (CBML) enhances the capabilities of the monolithic ML model. Extending monolithic ML approach, the paper presents deep-learning architectures, component development methods and evaluates their suitability for space exploration in building design. Results indicate that deep learning increases the performance of models over simple artificial neural network models. Methods such as transfer learning and Multi-Task Learning make the component development process more efficient. Testing the deep-learning model on 201 new design cases indicates that its cooling energy prediction (R2: 0.983) is similar to BPS, while errors for heating energy predictions (R2: 0.848) are higher than BPS. Higher heating energy prediction error can be resolved by collecting heating data using better design space sampling methods that cover the heating demand distribution effectively. Given that the accuracy of the deep-learning model for heating predictions can be increased, the major advantage of deep-learning models over BPS is their high computation speed. BPS required 1145 s to simulate 201 design cases. Using the deep-learning model, similar results can be obtained in 0.9 s. High computation speed makes deep-learning models suitable for design space exploration.
Uncertainty analysis in building energy assessment has become an active research field because a number of factors influencing energy use in buildings are inherently uncertain. This paper provides a systematic review on the latest research progress of uncertainty analysis in building energy assessment from four perspectives: uncertainty data sources, forward and inverse methods, application of uncertainty analysis, and available software. First, this paper describes the data sources of uncertainty in building performance analysis to provide a firm foundation for specifying variations of uncertainty factors affecting building energy. The next two sections focus on the forward and inverse methods. Forward uncertainty analysis propagates input uncertainty through building energy models to obtain variations of energy use, whereas inverse uncertainty analysis infers unknown input factors through building energy models based on energy data and prior information. For forward analysis, three types of approaches (Monte Carlo, non-sampling, and non-probabilistic) are discussed to provide sufficient choices of uncertainty methods depending on the purpose and specific application of a building project. For inverse analysis, recent research has concentrated more on Bayesian computation because Bayesian inverse methods can make full use of prior information on unknown variables. Fourth, several applications of uncertainty analysis in building energy assessment are discussed, including building stock analysis, HVAC system sizing, variations of sensitivity indicators, and optimization under uncertainty. Moreover, the software for uncertainty analysis is described to provide flexible computational environments for implementing uncertainty methods described in this review. This paper concludes with the trends and recommendations for further research to provide more convenient and robust uncertainty analysis of building energy. Uncertainty analysis has been ready to become the mainstream approach in building energy assessment although a number of issues still need to be addressed.
Both building performance analysis and multi-criteria performance optimisation often use deterministic simulations. Since many influencing parameters are generally inherently uncertain, this may lead to unreliable predictions of design impact. Therefore, this paper proposes a probabilistic analysis and design method to incorporate these uncertainties. The embedded Monte Carlo based uncertainty and sensitivity analyses investigate the output distributions. To greatly reduce computational efforts, meta-models can be incorporated, replacing the original model. Additionally, multi-layered sampling schemes are used to subject all design options to the same uncertainties and to check the validity of optimisation results for potential scenarios. Since reliability is a key aspect in this methodology, the paper also focuses on output convergence and method reliability.
To optimise both average performances and spread, effectiveness ɛ and robustness RP are introduced as output uncertainty indicators, inspired by robust design. Here, effectiveness is defined as the ability of the design option to optimise the performance, while robustness is defined as the ability to stabilise this performance for the entire range of input uncertainties.
The successive methodology steps are explained using a simplified application example.
Machine learning (ML) energy prediction models are useful to estimate the building energy demand with short response time, particularly important for developing energy-efficient building designs at an early stage of design. Component-based machine learning model (CBML) is the state-of-the-art development of ML energy prediction and useful to improve the generalizability of models. The concept of CBML is based on decomposing the design artefact in an engineering-related way and calculating intermediate parameters such as heat flows through building elements, followed by energy component at zone level and finally total energy demand at building level. Previous research showed that the method works but the improvement of the range of design and accuracy are still desirable to make the method more aligned to architectural design. Therefore, in this paper, we propose to enrich the training data with new building shapes and enhanced features typical for architectural design. For training each model of CBML, the data is collected by performing parametric energy simulations with three different building shapes and tested on fourth shape. A manual feature engineering approach is carried out by extracting useful features which have an influence on the target value. The accuracy of CBML is ascertained by coefficient-of-determination (R 2-values) and mean absolute percentage error (MAPE). The effect of enriching data with different building shape is studied by training ML model while increasing data sequentially and recording the improvements in the prediction accuracy. To study the effect of enhancing feature, CBML is developed with two types of features-raw and enhanced features and recording the change in the prediction accuracy. Furthermore, the influence of features is calculated using permutation importance to study the effect of additional features. The accuracy of total building energy model on new building shape improves from 5.18% to 3.14% (MAPE) and 0.9970 to 0.9988 (R 2) after enriching the data with several building shapes and enhancing the features.
The concept of multi-Level-of-Development (multi-LOD) modelling represents a flexible approach of information
management and compilation in building information modelling (BIM) on a set of consistent levels. From an
energy perspective during early architectural design, the refinement of design parameters by addition of information
allows a more precise prediction of building performance. The need for energy-efficient buildings
requires a designer to focus on the parameters in order of their ability to reduce uncertainty in energy performance
to prioritise energy relevant decisions. However, there is no method for assigning and prioritising information
for a particular level of multi-LOD. In this study, we performed a sensitivity analysis of energy models
to estimate the uncertainty caused by the design parameters in energy prediction. This study allows to rank the
design parameters in order of their influence on the energy prediction and determine the information required at
each level of multi-LOD approach. We have studied the parametric energy model of different building shapes
representing architectural design variation at the early design stage. A variance-based sensitivity analysis
method is used to calculate the uncertainty contribution of each design parameter. The three levels in the
uncertainty contribution by the group of parameters are identified which form the basis of information required
at each level of multi-LOD BIM approach. The first level includes geometrical parameters, the second level
includes technical specification and operational design parameters, and the third level includes window construction
and system efficiency parameters. These findings will be specifically useful in the development of a
multi-LOD approach to prioritise performance relevant decisions at early design phases.
Machine learning is increasingly being used to predict building performance. It replaces building performance simulation, and is used for data analytics. Major benefits include the simplification of prediction models and a dramatic reduction in computation times. However, the monolithic whole-building models suffer from a limited transfer of models and their data to other contexts. This imposes a vital limitation on the application of machine learning in building design. In this paper, we present a component-based approach that develops machine learning models not only for a parameterized whole building design, but for parameterized components of the design as well. Two decomposition levels, namely construction level components (wall, windows, floors, roof, etc.), and zone-level components, are examined. Results in test cases show that, depending on how far the cases deviate from the training case and its data, high prediction quality may be achieved with errors as low as 3.7% for cooling and 3.9% for heating.
Approximately half of the annual global energy supply is consumed in constructing, operating, and maintaining buildings. Because most of this energy comes from fossil fuels, it also contributes greatly to annual carbon emissions. When constructing a building, embodied energy is consumed through construction materials, building products, and construction processes along with any transportation, administration, and management involved. Operating energy is used in space conditioning, heating, lighting, and powering building appliances. In order to effectively reduce the carbon footprint of buildings, a comprehensive reduction in both embodied and operating energy is needed. Studies so far have focused on reducing either embodied or operating energy in isolation without realizing the trade-off that exists between them. Also, building energy research has concentrated more on operating energy than embodied energy, and as a result, the operating energy of buildings is gradually decreasing. Due to a variety of issues, however, few efforts have been undertaken to comprehensively minimize embodied energy.
Building form does influence energy consumption. Designing low-energy architecture to minimize energy consumption requires thoughtful articulation of the shape and form of a building. The Architect's decision-making for more energy efficient building form is often based on rules of thumb. Historically, the rule of thumb regarding passive solar building design suggests that form and orientation matter to overall energy performance. The question of how much impact does form have varies between project to project, due to climate, location, and building size. However, evaluation of energy performance specifically relating to building form is difficult to quantify because of the large solution space, but nonetheless important to understand.
BIM (building information model) enables information sharing and reuse for interoperability between prevalent software tools in the AEC (Architecture, Engineering, and Construction) industry. Although a BIM based energy simulation tool can reduce costs and time required for building energy simulation work, no practical interface between CAD tools and dynamic energy analysis tools has been developed so far. With this in mind, this study suggests two approaches (Full automated interface (FAI) and semi automated interface (SAI)) enabling information transition from CAD tools (e.g., IFC) to EnergyPlus input file, IDF. FAI, if ideally developed, can convert IFC to IDF based on the use of pre-defined defaults without requiring human intervention. In contrast, SAI converts geometry information drawn from IFC to IDF and then require human data entry for uncertain simulation inputs. For this study, a library building was chosen and space boundary generated from ArchiCAD 13 was employed for geometry mapping. The Morris method, one of sensitivity analysis methods, was used for identifying significant inputs. In FAI and SAI, dominant inputs, out of the Morris method, were identified for Monte Carlo simulation to quantify probabilistic simulation outputs. In the paper, FAI and SAI simulation results are cross-compared, and pros and cons of FAI and SAI are discussed.
Objectives
IntroductionThe Chi-Square Goodness-of-Fit TestThe Chi-Square Test for IndependenceThe Fisher Exact TestExamples from the LiteratureSummaryPractice QuestionsSolutions to Practice Questions