Urban Building Energy Prediction at Community Scale

Predictive models for urban building energy use have been the focus of much research in recent years, especially using data-driven techniques. However, these models still need to address recognized challenges, such as employing sufficient energy use data in spatial and temporal scales and accounting for interbuilding effects. In this regard, several typical data-driven predictive models for urban building energy use were proposed in this capstone to reduce the large data requirements and improve the prediction accuracy. Using a dataset of four years of electricity consumption by public buildings in Jianhu City, a county-level city in Jiangsu Province, China, and data on the corresponding building morphological parameters, this project compares the predictive performance of these models under different algorithms. The results suggest that a building network based on building morphological similarity can improve the overall performance of energy consumption prediction models for individual buildings in an urban context. This building network can also obtain relatively reliable energy consumption prediction results in the absence of historical energy consumption data of the target building. The project also reveals that the data-driven models can accurately predict total building consumption in a region when historical energy consumption of some buildings is not available. This study provides more comprehensive references and improved accuracy and robustness of urban building energy demand prediction, resulting in potential solutions reduced data requirements of urban energy models.