The applicability of grey-box performance-based modelling techniques in existing office buildings in the UK

本博士研究基于灰箱性能的建模技术在描述英国现有办公楼的热力动力学方面的适用性。该项目旨在回答灰箱模型能在多大程度上预测现有办公楼的内部空气温度和热损失，以评估改造方案和控制策略。以一栋超节能被动式住宅办公楼为例，以最小的计算复杂度和最小的计算成本确定了最佳灰箱模型结构，并提供了最高的精度。随着楼宇自动化系统(BAS)和物联网(IoT)在建筑物中的广泛应用，传感器和其他来源不断地收集与建筑物及其设备的功能有关的大量测量数据。这为创建用于建筑控制和运营的数据驱动模型提供了大量机会。数据驱动建模，也称为基于性能的建模或反向方法，是基于建筑物入驻后的测量数据。这些模型反映了实际的建筑热动态，并提供了更准确的建筑热响应预测。因此，基于性能的模型可用于空间供暖和制冷的基于模型的控制、机械系统的故障检测、降低运行能耗的改造评估、调峰调峰和性能监测。此外，如果建模过程还可以利用建筑物运行期间记录的运行测量，则可以减少建筑性能模拟(BPS)中的不确定性。这些不确定性是由各种各样的动态、随机和概率因素造成的，如建筑几何、材料属性、暖通空调系统、居住者行为、设备、使用进度甚至天气数据，以及在现有建筑的情况下，可获得的信息较少的情况下的增加。尽管近几十年来，人工神经网络(ANN)等机器学习技术已被广泛应用于建筑能耗预测，但这些技术存在一些缺点，包括对数据质量的高要求、缺乏可解释性和高计算要求。这种方法还会在不同建筑之间产生概括性较低的模型，这使得模型很难在建筑物之间进行比较。比较建筑物的热性能对于能源消耗分类是有意义的。灰箱模型是一种数据驱动的模型，它保留了一些物理意义，而它们的参数是使用测量数据校准的。因此，灰箱模型比机器学习方法更具解释性，计算效率更高。

This Ph.D. investigates the applicability of grey-box performance-based modelling techniques to describe the thermal dynamics of an existing office building in the UK. The project aims to answer the question of how well grey-box models can predict the internal air temperature and heat loss of an existing office building, in order to evaluate retrofit options and control strategies. An ultra energy-efficient Passive House office building is studied as the case study to identify the best grey-box model structure in terms of providing the highest accuracy with minimum complexity and computational cost. The type, amount and quality of data coming from sensors of a Building Management System (BMS) and the level of detail (LOD) required to inform the model parameters for the office building will be explored.With the wide adoption of building automation systems (BAS) and the Internet of Things (IoT) in buildings, numerous measurements pertaining to the functioning of the buildings and their equipment are continuously gathered by sensors and other sources. This offers numerous chances to create data-driven models for building control and operation. Data-driven modelling, also known as performance-based modelling or inverse approach, is based on measured data after buildings are occupied. These models reflect the actual building thermal dynamics and provide more accurate predictions of building thermal responses. Therefore, performance-based models can be used in model-based control of space heating and cooling, fault detection of mechanical systems, retrofit evaluation for reducing operation energy consumption, shifting and shaving peak demand, and performance monitoring. Moreover, the uncertainties in Building Performance Simulation (BPS) can be reduced if the modelling process can also make use of operational measurements, recorded during building's operation. These uncertainties can result from of a wide range of dynamic, stochastic, and probabilistic elements such as building geometry, material properties, HVAC systems, occupant behaviour, appliance, use scheduling and even weather data, and increase in the case of existing buildings, where less information is available.Although machine learning techniques, such as Artificial Neural Networks (ANN) have been extensively adopted for building energy use prediction in recent decades, despite their high prediction accuracy, these techniques have some drawbacks, including a high demand for data quality, lack of interpretability and intense computational requirements. This approach also produces models with low generalisation between different buildings, which makes building-to-building comparisons of models difficult. Comparing the thermal behaviour of buildings can be of interest for the purposes of energy consumption classification. Grey-box models are a type of data-driven models that retain some physical meaning, while their parameters are calibrated using measured data. Therefore, the grey-box model is more interpretable than machine learning approach and is more computationally efficient.