Research on Urban Building Energy Consumption Simulation Based on a Three-Dimensional Geographic Information System

Abstract

This study develops a building energy consumption proxy model integrating 3D GIS, energy simulation software, and machine learning for large-scale analysis. Using Shanhe Bay Valley as a case study, geometric building data was first obtained through water economy micromaps. Energy models for sample buildings were created using SketchUp, OpenStudio, and EnergyPlus to generate machine-learning datasets. Four machine learning methods were evaluated via R language, identifying the MARS model (degree=2, nprun=10) as optimal. The model predicted annual operational carbon emissions at 74.40 kg/(m²·a), with heating, cooling, and electricity energy densities of 0.252 GJ/m², 0.175 GJ/m², and 0.112 GJ/m² respectively. Sensitivity analysis revealed five key energy drivers: per capita occupied area (most influential), ventilation frequency, solar heat gain coefficient, lighting power density, and equipment power density. These findings demonstrate the model's effectiveness in quantifying energy performance patterns across urban building clusters while identifying key optimization parameters for sustainable design.

Keywords: Urban scale, Machine Learning, Software simulation, Energy consumption proxy model, Sensitivity analysis