Description
Over recent decades, the increase in global energy consumption has led to negative environmental impacts and has become a major concern hindering the development of human society. Since 2007, the Chinese government has implemented a large-scale building energy efficiency retrofit policy. However, in practice, planning optimal retrofits for buildings demands substantial professional expertise and involves a time-consuming and labour-intensive simulation process. To support decision-making for retrofit projects, this thesis presents a novel methodology for developing a data-driven energy prediction model.The research unfolds in four interconnected stages. First, an automated framework for architectural element recognition and semantic retrieval is developed based on existing floor plans of ageing residential buildings, enabling the automatic generation of baseline Building Energy Models (BEMs). In the second stage, these baseline models are enriched with thermal material properties and occupant behaviour parameters, defined in accordance with Chinese building energy efficiency standards and literature. Large-scale parametric energy simulations are then carried out to produce a comprehensive residential energy consumption dataset tailored to the Chinese housing context. In the third stage, a multi-modal deep learning model, FusionGAT, is developed. This framework represents each multi-family residential building as a graph, where each household unit is treated as a node and thermal interactions between units are explicitly modelled through edge features. FusionGAT employs a Convolutional Neural Network to extract geometric features from floor plans and a Transformer-based encoder to capture occupant behaviour patterns from daily usage schedules. The fused multi-modal features are subsequently processed by a Graph Attention Network to predict energy consumption at the household level. The model is first pre-trained on the comprehensive simulation dataset and then fine-tuned using real-world data collected from 120 households, enhancing its adaptability and predictive accuracy in practical settings. Finally, the calibrated FusionGAT model is applied to a case study of a typical residential community to rapidly evaluate energy performance under various retrofit scenarios. By combining predicted energy outcomes with detailed retrofit cost assessments, a multi-objective optimisation is conducted to identify and analyse the most effective retrofit strategies. The results provide data-driven insights to support stakeholders in China’s residential building sector in making informed decisions that balance energy savings with economic feasibility
| Period | 15 Aug 2025 → 20 Dec 2025 |
|---|---|
| Examination held at |
|
| Degree of Recognition | International |