Bayesian processes for calibration of building performance simulation models and optimisation of building energy systems design

自2015年达成具有里程碑意义的《巴黎气候变化协定》以来，国际组织、各国政府和全球行业已开始做出新的努力，制定政策路径、激励计划和技术选择，以立即减少由建筑引起的温室气体排放。特别是在欧洲、北美和东亚等监管严格的建筑市场，建筑规范和标准的重新制定越来越多地着眼于使用建筑性能模拟(BPS)工具来促进决策和合规，以进行节能、低碳的建筑设计和翻新。随着这一过程的进行，在现实世界中，基于BPS的决策的成功仍然存在两个持久的障碍：1)经常遇到的性能差距，表现为专家驱动的建筑能耗模型预测与实际测量之间的差异，以及2)缺乏廉价和可获得的BPS模型校准技术。这个项目的长期目标是改变传统的建筑设计做法，通过使用BPS工具进行决策的建筑设计问题的预测不确定性的明确表示。具体地说，该项目建议研究、开发和验证：1)使用贝叶斯定理校准瞬时BPS模型的新方法，2)使用贝叶斯过程对建筑设计进行随机优化的新方法，以及3)使用贝叶斯过程对未来建筑性能进行代理建模的新方法。为了促进项目的长期目标，将按顺序实现以下短期目标：(1)开发并执行一种方法，以吸收温哥华大都会地区至少18个候选商业、住宅和机构建筑的建筑施工信息和性能基准数据；(2)开发并执行专家启发过程，以征求与每个建筑的建筑信息相关的任何先前不确定性知识；(3)建立一个最佳的方法来处理特定BPS子过程中的不确定性，例如入住率预测、建筑能源需求模拟和能源供应系统建模(4)系统地测试和验证每个候选建筑的BPS模型的贝叶斯校准的新方法；(5)对贝叶斯BPS代理模型应用于建筑设计的随机优化和/或对未来建筑能源和环境状况的稳健预测的适用性进行实验研究。这项工作的总体成果将是验证和传播一种新的数据辅助模拟方案，用于在普遍存在的技术和经济不确定性下预测和优化加拿大建筑物的能源消耗。

Since the landmark Paris Agreement on Climate Change was struck in 2015, international organisations, national governments, and global industry have started new efforts at developing policy pathways, incentive schemes, and technology options for an immediate reduction in building-attributed greenhouse gas emissions. Particularly in the highly-regulated building markets of Europe, North America, and East Asia, redevelopments of building codes and standards have increasingly looked towards the use of building performance simulation (BPS) tools to facilitate decision-making and code-compliance for energy-efficient, low-carbon building design and retrofits. As this is ongoing, there remains two persistent barriers to the success of BPS-based decision-making in the real world: 1) the frequently-encountered performance gap represented by the discrepancy between expert-driven model predictions of building energy use and actual measurements, and 2) the lack of inexpensive and accessible techniques for BPS model calibration. The long-term objective of this project is to transform conventional building design practises by enabling the explicit representation of prediction uncertainty in building design problems that use BPS tools for decision-making. Specifically, the project proposes to research, develop, and validate: 1) a new approach to calibration of transient BPS models using Bayes theorem, 2) a new approach to stochastic optimisation of building design using Bayesian processes, and 3) a new approach to surrogate modelling of future building performance using Bayesian processes. To facilitate the project's long-term objective, the following short-term objectives will be delivered in sequence: (1) develop and execute an approach to assimilating building construction information and performance benchmarking data from at least 18 candidate commercial, residential, and institutional buildings in the Metro Vancouver region; (2) develop and execute an expert elicitation process to solicit any prior knowledge of uncertainty related to each building's construction information; (3) establish a best-practise approach to handling uncertainty within specific BPS sub-processes, such as occupancy prediction, building energy demand simulation, and energy supply systems modelling (4) systematically test and validate a new approach to Bayesian calibration of BPS models for each candidate building; (5) undertake an experimental investigation of the suitability of Bayesian BPS surrogate models to be applied for stochastic optimisation of building design and/or robust prediction of future building energy and environmental conditions. The overall outcome of the work will be the validation and dissemination of a new data-assisted simulation schema for predicting and optimising the energy consumption of buildings in Canada under prevailing technical and economic uncertainties.