Gas Prediction of Sino-Singapore Tianjin Eco-City Based on ARMA Model

Yizhi Xu, Ling Peng, Li Xiang, Gaosheng Li, Shaolong Cui2015 · DOI

Abstract

With the development of Chinese economic growth, energy consumption is growing. Among them, gas is an important energy source for urban living, and its consumption has increased year by year. In this paper, based on a data aggregation platform and residents' daily natural gas consumption data in the Sino-Singapore Tianjin Eco-City (SSTEC), a time series prediction model, ARMA, was used to predict daily gas consumption by considering the time series correlation of the data.

Introduction

As China undergoes rapid urbanization, the national energy mix is shifting toward cleaner sources like natural gas. Managing this transition requires precise forecasting to ensure steady supply and minimize waste. The Sino-Singapore Tianjin Eco-City (SSTEC) serves as a "Smart City" pilot project, generating high-resolution data through its integrated platforms. This study leverages data from 22,159 households across 15 residential districts (including Hong Shuwan and Mei Linyuan) collected between January 1 and November 30, 2014. By analyzing 300 days of historical consumption, the research aims to provide actionable short-term forecasts for urban energy planning.

Methodology

The researchers employed the Auto-Regressive Moving Average (ARMA) model, a foundational tool in statistical time-series analysis. The implementation followed a structured six-step pipeline to ensure data integrity and model accuracy.

The ARMA modeling and prediction pipeline used in the study.
  1. Data Preprocessing: Raw consumption data was converted into a zero-mean, stationary sequence to stabilize statistical properties over time.
  2. Model Identification: Autocorrelation and partial autocorrelation functions were analyzed to determine the optimal model structure.
  3. Parameter Estimation: Coefficients were calculated to define the relationship between past and future values.
  4. Model Diagnosis: Using the Akaike Information Criterion (AIC), the team identified ARMA(5,3) as the most efficient model.
  5. Forecast: The model was used to predict consumption for a 7-day test period.
  6. Error Analysis: Predictions were validated against actual measured values to assess reliability.

Key Results and Insights

The ARMA(5,3) model demonstrated high precision in both fitting and prediction phases. During the training phase, the model achieved a Root Mean Square Error (RMSE) of 0.1019. Notably, the prediction accuracy improved during the 7-day test period, with the RMSE dropping to 0.0749.

Comparison between the original gas consumption sequence and the forecasting sequence.

Analysis of the data revealed that residential gas consumption is remarkably stable between adjacent days, making it highly suitable for ARMA modeling. However, clear seasonal cycles were observed, with consumption trending upward during the fall and winter months. While the model occasionally showed higher error on the first day of prediction (~14.7%), it stabilized quickly, with relative errors as low as 0.5% by the third day. This suggests that even standard statistical models can provide high-precision results for complex urban utilities without the need for computationally intensive deep learning.

Conclusion and Future Work

While the ARMA model proved effective for temporal forecasting, it currently lacks spatial awareness. Future research will focus on incorporating spatial correlations—how usage in one district affects another—and expanding the system to include water and thermal energy. The ultimate goal is to develop a comprehensive, multi-utility prediction system for holistic resource management across the entire Eco-City.

Summary updated

Abstract

With the development of Chinese economic growth, energy consumption is growing. Among them, the gas is an important energy source of urban living, and its consumption has increased year by year. In this paper, based on the data aggregation platform, and residents daily natural gas consumption data in SSTEC(Sino-Singapore Tianjin Eco-City)[1], considering the time series correlation of data, a time series prediction model ARMA was used to predict the daily consumption of gas in SSTEC.
BibTeX
@misc{xu2015gas,
  author = {Yizhi Xu and Ling Peng and Li Xiang and Gaosheng Li and Shaolong Cui},
  title = {{Gas Prediction of Sino-Singapore Tianjin Eco-City Based on ARMA Model}},
  year = {2015},
  doi = {10.2991/icaees-15.2015.261},
}
Figure