Engineering-based reduced-order modelling of districts in the context of heuristic life cycle assessment

由于雄心勃勃的全球和国家气候保护目标，生命周期评估(LCA)方法正迅速变得重要起来。在建筑部门，对温室气体排放量和更多污染物的地区规模估计对于广泛确定脱碳措施是必要的。然而，对所有区域生命周期阶段的考虑意味着输入和输出参数的高度粒度和敏感度。缺乏生命周期清单(LCI)数据和大规模确定建筑能源需求的高计算工作量加剧了这一问题。为了解决这些问题，原型和低阶建筑模型已经被开发出来。前者旨在对相似度较高的建筑物进行聚类，而后者旨在减少所需输入参数的总量。然而，原型的静态性质和所有生命周期阶段输入参数的固有波动性强调了可持续调整的模型的必要性。此外，在广泛的地区范围内确定污染物，需要在新建和翻新时考虑住宅和非住宅建筑的一般方法。这种方法应该能够借助小范围的输入参数来估计排放量，以减少计算工作量。因此，该项目的目的是开发一种通用的自上而下的城市建模方法，用于环境指标的启发式预测。应使用大量的内源和外源数据，以探索目前尚不清楚的建模和模拟输入与影响评估输出之间的统计显著相关性。对于数据内插和丰富，将使用合成工具。将使用数据预处理和数据挖掘来准备收集的数据作为不同机器学习方法的训练和测试集。这些方法将从准确性、运行时间和数据需求方面进行评估，以确定性能最好的建模方法。为了提高用于生命周期评价的高度详细的建筑模型的普遍可用性，应通过使用收集的数据来改进用于对CityGML建筑模型进行内插、变换和丰富的现有综合工具。开发的机器学习模型将公开用于持续适应，并产生比静态原型方法更准确的区域LCA结果。

The methodology of life cycle assessment (LCA) has been gaining importance rapidly due to ambitious global and national climate protection goals. In the building sector, district-scale estimations of greenhouse gas emissions and further pollutants are necessary for the broad determination of decarbonisation measures. Yet, the consideration of all district life cycle phases implies a high granularity and sensitivity of input and output parameters. This is exacerbated by a lack of life cycle inventory (LCI) data and high computational effort for the large-scale determination of building energy demands. To tackle these issues, archetypes and low-order building models have been developed. The former is intended to cluster buildings with a high degree of similarity, while the latter is meant to reduce the overall amount of necessary input parameters. However, the static nature of archetypes and inherent volatility of input parameters along all life cycle phases emphasise the necessity of a model that can be adapted continuously. Furthermore, the broad district-scale determination of pollutants requires a generic approach for the consideration of residential as well as non-residential buildings in new construction and refurbishment. This approach should make it possible to estimate emissions by aid of a small range of input parameters to reduce the computational effort. Thus, the aim of this project is to develop a generic top-down urban modelling approach for the heuristic prediction of environmental indicators. A broad bandwidth of endogenous and exogenous data shall be used to explore statistically significant correlations currently unknown between modelling and simulation input on the one hand, and impact assessment output on the other hand. For data interpolation and enrichment, synthetisation tools will be employed. Data pre-processing and data mining will be used to prepare the collected data as training and testing sets for different machine learning methods. These methods will be evaluated in terms of accuracy, runtime and data demand to identify the best-performing modelling approaches. To increase the general availability of highly detailed building models for LCA purposes, pre-existing synthetisation tools for interpolation, transformation and enrichment of CityGML building models shall be advanced by the use of the collected data. The developed machine learning models will be made publicly available for the continuous adaptation and to produce more accurate district LCA results than a static archetype approach.