Development of a surrogate model for interactive early-stage net-zero building design
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The Net-Zero Navigator is a collaborative effort to leverage surrogate modeling for use in an opensource, online early stage building energy modeling and decision-making tool, formed among industry, academic and government stakeholders in Canada. The surrogate modelling underpinning Net Zero Navigator leverages machine learning techniques to train statistical meta-models based on BPS input and output data, emulating high-fidelity building simulation and providing an engine for rapid design performance estimates. The platform is built on a foundation of existing tools (EnergyPlus, TensorFlow, KERAS API) as well as the codebase of the Building and Energy Simulation, Optimization and Surrogate-modelling (BESOS) platform. Overall, the Net-Zero Navigator platform supports fast, interactive concept-stage building design, at a stage where decisions have a proportionally greater impact on downstream performance, and where flexibility and speed are important. To further disseminate the knowledge gained through development of Net Zero Navigator, this study dissects the architecture and hyperparameters that define the artificial neural networks, demonstrating importance for specific aspects of the building energy estimation through an experimental procedure aligned with hyperparameter optimization methods found in machine learning best practice. This methodology involves hundreds of thousands of EnergyPlus simulation runs and equivalent numbers of neural network configurations.High-performance building design requires rapid iteration involving many highly integrated decisions, with diminishing potential for impact as the design process progresses. Existing high-fidelity simulation approaches have limited ability to deliver fast, integrated design iterations, which can compromise the ability to achieve net-zero energy in new buildings.In most cases, these tools rely on generating direct representations of the building explicitly, using detail-oriented, physics-based modelling engines. This limits the computational speed and flexibility, as well as the ability to explore the complex problem space effectively. Attempts to simplify the model input process in these tools may also contribute to errors where the underlying engine cannot adaptively and automatically account for complex interactions that can result.Further research has shown that machine learning methods are an area of study with potential to connect the needs during early concept design and planning with a viable toolset for use by practitioners and researchers. Specifically, surrogate methods have been shown to produce robust, reliable building energy analysis results while greatly reducing computational time and input complexity by the user. However, there is trade-off potential between absolute accuracy and flexibility, breadth and speed of design space exploration that must also be examined.As a nascent field of study, surrogate modelling in the buildings domain faces many challenges, including loss of interpretability and amplification of errors due to poorly framed problems. However, as a new addition to building design toolset, with novel advantages and potential for development, surrogate modelling can be used to supplement traditional analytical and high-fidelity techniques. The Net Zero Navigator project was developed to help resolve these limitations.The results of this study highlight some of the computational benefits, the scale of modeling uncertainty introduced, and the relationships between energy model complexity and surrogate model configuration. Furthermore, this study provides a basis to discuss the benefits of Net Zero Navigator and surrogate modeling for building energy simulation, leveraging the multi-disciplinary development efforts to suggest best practice and promising directions for future work.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it