National-Scale UBEM with Intrinsic Degree-Hour (IDH) Method in Republic of Korea [UBEM-IDH-ROK]

Introduction

This repository contains Python scripts and visualized data for the Urban Building Energy Modeling (UBEM) project targeting the Republic of Korea.

Utilizing the Intrinsic Degree-Hour (IDH) method, this project analyzes the energy performance of approximately 1.6 million buildings nationwide. Unlike traditional degree-day methods that use fixed base temperatures (e.g., 18.3°C), the IDH approach identifies the specific Optimal Base Temperature ($T_{b,opt}$) for each individual building based on its monthly energy consumption patterns.

Key Features

1. Minimal Data Requirement

• Overcoming Data Scarcity: Analyzes building energy performance using only two variables: Monthly Energy Bill and Outdoor Temperature.
• Accessibility: Eliminates the need for complex physical datasets (e.g., 3D geometry, envelope insulation details) which are often unavailable at the urban scale.

2. Physics-Informed Interpretability

• Beyond Black-Box: Unlike simple statistical correlations, this method derives meaningful physical parameters:
  • $T_{b,opt}$: The unique 'Balance Point Temperature' for each building.
  • $U''$: The overall heat loss coefficient (Insulation performance).
• Behavioral Insight: Captures actual occupant behavior (e.g., heating preferences) rather than relying on standard design assumptions.

3. Scalability to National Level

• Massive Coverage: Successfully processed 1.2+ million buildings across South Korea by leveraging the computational efficiency of the algorithm.
• Spatio-Temporal Integration: Integrated local weather data from 146 stations to reflect precise local climate conditions for every administrative district (SIGUNGU).

---

Methodology: The IDH Method

The core concept is to find the unique Balance Point Temperature(IDH) for each building where heating/cooling is not required. (Ahn et al., 2025)

1. Governing Equation

The monthly energy consumption ($E_m$) is modeled as:

$$E_m = U'' \times IDH(T_{b,opt}) + E_{base}$$

Where:

• $T_{b,opt}$: Optimal Base Temperature (Search Range: $10^\circ C \sim 30^\circ C$)
• $U''$: Total Energy Performance Coefficient (Slope) indicating insulation/efficiency.
• $IDH$: Intrinsic Degree-Hours calculated from hourly weather data.
• $E_{base}$: Base Load (Intercept), representing weather-independent energy usage.

2. Optimization Algorithm

The algorithm iterates through a range of potential base temperatures ($10^\circ C$ to $30^\circ C$) for each building to find the specific $T_b$ that maximizes the Coefficient of Determination ($R^2$).

---

Repository Structure

├── data/
│   ├── raw/                # Original Energy & Weather Data (Private)
│   └── processed/          # Cleaned datasets & IDH lookup tables (Private)
├── src/
│   ├── code_dataprep.ipynb      # Scripts for loading and merging data
│   ├── code_regression.ipynb   # Calculate IDH / regression &optimization
│   └── code_visualize.ipynb  # Review and visualize final data
├── results/
│   ├── UBEM_Result_Final.csv  # Final output (Private)
│   └── figures/            # Visualization plots
└── README.md

Outputs & Results

The final analysis generates a CSV file containing IDH model parameters for each building.

1. Result File Structure (UBEM_Result_Final.csv)

Column	Description	Data Type
PNU	Unique Building Identifier (19-digit code)	String
STN_ID	Weather Station ID (KMA)	Integer
Tb_opt	Optimal Base Temperature ($^\circ C$)	Float
U_coeff	Total Energy Performance Coefficient (Slope)	Float
E_base	Base Energy Load (Intercept)	Float
R2	Model Accuracy Score ($0.0 \sim 1.0$)	Float
Sample_N	Number of monthly data points used	Integer

2. Result

First Analysis (Only R2>0)

>> Original Buildings : 1,233,505
>> Valid Buildings : 1,226,227 (R2 > 0)
============================================================
[IDH Modeling First Report]
============================================================
1. Success Rate : 99.4% (Valid Buildings / Every Building)
2. Model Accuracy Average : 0.5934 (R2 Score)
3. Highly Trusted Buildings : 497,684 (40.6% is R2 >= 0.7)
------------------------------------------------------------
4. The Most Frequent IDH : 10°C
5. Average IDH : 18.38°C
------------------------------------------------------------
6. Average Thermal Insulation Performance (U) : 1.32
============================================================

Second Analysis (Only 10<IDH<29 & 0<U<Top5%)

>> First Valid Buildings : 1,226,227
>> Second Valid Buildings : 845,952
============================================================
[IDH Modeling Second Report]
============================================================
1. Success Rate : 69.0% (2nd Valid Buildings / 1st Valid Buildings)
2. Refined Model Accuracy Average : 0.6254 (R2 Score)
3. Highly Trusted Buildings : 385,920 (45.6% is R2 >= 0.7)
------------------------------------------------------------
4. The Most Frequent IDH : 21°C
5. Average IDH : 18.50°C
------------------------------------------------------------
6. Average Thermal Insulation Performance (U) : 0.29
============================================================

• Key Finding: After refinement, the most frequent base temperature shifted to 21°C. This indicates that Korean buildings tend to start heating at higher outdoor temperatures compared to the conventional standard of 18.3°C (65°F).

Limitations & Future Work

Limitations

• Building Usage Classification: The current analysis applies a unified model across all building types (Residential, Commercial, Industrial, etc.) without distinction. This may include non-heated spaces (e.g., warehouses), leading to statistical outliers.
• Spatial Granularity: Due to the massive dataset size, the analysis focuses on administrative districts (SIGUNGU) rather than precise geospatial coordinates (Latitude/Longitude), limiting detailed spatial interpretation.

Future Plans

• Targeted Spatial Analysis: Future research will narrow the spatial scope to specific cities (e.g., Seoul, Busan) to conduct high-resolution spatial analysis.
• Multidimensional Integration:
  • Building Registers: Integrating detailed building ledgers to classify energy patterns by usage (Apartments, Offices, Retail).
  • Socio-Economic Factors: Analyzing correlations with regional income levels, population density, and building age.
  • Urban Context: Considering micro-climate factors such as the Urban Heat Island (UHI) effect.

---

Requirements

To run this project, you need the following environment:

• Language: Python 3.8+
• Libraries: pip install pandas numpy scipy scikit-learn geopandas tqdm requests matplotlib seaborn

Data Sources

This project utilizes official public datasets from the South Korean government. Note: Due to data privacy and redistribution policies, the raw datasets are not included in this repository.

1. Building Energy Consumption Data

• Source: Architecture HUB (건축데이터 민간개방 시스템)
• URL: https://open.eais.go.kr/
• Description: Monthly electricity and gas consumption data for individual buildings, provided by the Ministry of Land, Infrastructure and Transport (MOLIT).

2. Meteorological Data

• Source: Korea Meteorological Administration (KMA) Weather Data Service
• URL: https://data.kma.go.kr/
• Description: Hourly outdoor temperature data collected from 146 Synoptic Weather Observation Systems (ASOS) nationwide.

3. Geospatial Data (Administrative Divisions)

• Source: Statistics Korea (KOSTAT) - SGIS
• Repository: southkorea/southkorea-maps
• Description: GeoJSON data for South Korean municipal boundaries (2013 version), originally provided by the Statistical Geographic Information Service (SGIS) and simplified by the open-source community.

Data Privacy & License

Code License

• This project is licensed under the MIT License.
• The methodology of the algorithm is based on Ki Uhn Ahn, Kim, Deuk-Woo and Kim Hye-Gi. (2025). Intrinsic Degree Hour (IDH) Methodology for Data-Driven Urban Building Energy Modeling (UBEM). 한국건축친환경설비학회 논문집, 19(4), 180-190.

Data Disclaimer

• The raw building energy data and temperature data are confidential and were obtained for specific research purposes.
• Users wishing to replicate this study must obtain their own access permissions from the relevant data providers.