Design and Application of an Energy Management System Based on Artificial Intelligence Technology

Hong-Ye Lin, Xuemei Bai, Chun Li, Shuguang Xu, Haibin Xu, Zne-Jung Lee, Yun Lin, Qing Zhou, Jinhai Cai2025 · DOI

TL;DR

This paper presents an AI-powered energy management system for buildings, focusing on predicting chilled water return temperatures to optimize energy usage. The system integrates sensor data, cloud-based processing, machine learning for prediction, and a large-screen visualization interface, demonstrating the effectiveness of Support Vector Regression (SVR) for this task.

What Problem Does This Paper Address?

With China's rapid economic growth comes a significant increase in energy consumption and resource scarcity, driving the need for sustainable development and effective energy management. Intelligent buildings offer a prime opportunity to implement energy-saving strategies. However, efficiently monitoring, analyzing, and predicting energy usage in real-time remains a challenge, especially for small and medium-sized enterprises due to the high cost and complexity of existing solutions. This paper aims to bridge this gap by developing an accessible and efficient AI-driven energy management platform.

Methods

The core of the system is an AI-powered energy management platform with a modular architecture. It comprises:

  • Sensors and Data Acquisition: Real-time collection of energy consumption data from various building systems.
  • Cloud Platform: A centralized system for data storage, processing, and analysis. This includes a custom backend management system built using the FastAdmin framework, leveraging PHP and MySQL.
  • Machine Learning for Prediction: The study explores several regression algorithms to predict the chilled water energy meter return temperature of central air-conditioning systems. These include:
  • Decision Tree Regression
    • Support Vector Regression (SVR)
    • K-Nearest Neighbors (KNN) Regression
    • Extreme Gradient Boosting (XGBoost) Regression
  • Data Preprocessing: Data is scaled to a [0,1] range, and then split into training and testing sets.
  • Result Evaluation and Alarm: Predicted values are compared against actual temperatures. If the difference exceeds a set threshold, an alarm is triggered.
  • Large-Screen Display: A visualization interface using the Pyecharts library provides an intuitive overview of energy data, including real-time temperature trends, system status, and geographical distribution of energy usage.

Datasets

The study utilized a temperature dataset provided by a company in Xiamen, China. This dataset contained 18 attributes, all continuous variables. 17 of these attributes were used as features to predict the chilled water energy meter return temperature. The data was recorded every ten minutes.

Key Insights and Ideas

  • Holistic System Design: The paper emphasizes a complete system lifecycle, from data collection and cloud platform construction to machine learning analysis and visualization, offering a comprehensive solution.
  • SVR's Strength in Nonlinearity: The study highlights SVR's advantage in handling complex, nonlinear relationships within the data, which is crucial for accurately predicting energy consumption patterns.
  • Importance of Visualization: The use of a large-screen display with Pyecharts is key to making complex energy data easily understandable for managers, facilitating informed decision-making.
  • Practical Application for SMEs: By developing a system that is potentially more accessible than existing high-cost solutions, the research aims to enable wider adoption of AI in energy management.

Key Results

The performance of the machine learning algorithms was evaluated using Mean Squared Error (MSE). The results were as follows:

  • Support Vector Regression (SVR): Achieved the lowest MSE of 0.09, indicating the best prediction performance.
  • Extreme Gradient Boosting (XGBoost): MSE of 0.32.
  • Decision Tree Regression: MSE of 0.36.
  • K-Nearest Neighbors (KNN) Regression: Had the highest MSE of 0.57, showing poorer performance compared to the other algorithms.

The paper concludes that SVR, XGBoost, and Decision Tree regression performed well, with SVR being the top performer.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • Limited Scope: The current model is primarily applicable to central air-conditioning systems. Further testing and adaptation are needed for other energy systems.
  • Data Requirements: More data is required to further improve the model's accuracy and performance.
  • Enhanced Functionality: Future work should focus on improving intelligent data acquisition and analysis capabilities, as well as incorporating diverse types of energy management.
  • Hardware Integration: The use of smart sensors and IoT devices for real-time monitoring and iterative updates of smart algorithms is recommended for enhanced system capabilities.

Summary updated

Abstract

In response to the increasingly severe energy consumption problem and to promote energy saving and emission reduction, this study aims to design and apply an energy management system platform based on artificial intelligence (AI) technology. The system adopts sensor technology and data acquisition equipment to monitor various types of energy consumption in buildings in real time, efficiently process and predict these data through machine learning algorithms, and finally visualize the results. The system is functionally complete, completing the process from data collection to visualization, the cloud platform’s construction, and finally a full energy management platform. Various machine learning methods are applied to energy management by predicting the chilled water energy meter return temperature of the central air-conditioning system and comparing its performance. Among the various types of regression algorithms, the mean-square error (MSE) of decision tree regression is 0.36, the MSE of support vector regression (SVR) is 0.09, the MSE of K-nearest neighbor (KNN) regression is 0.57, and the MSE of extreme gradient boosting (XGBoost) regression is 0.32. The SVR, the XGBoost regression, and the decision tree regression perform better in various indices.
BibTeX
@misc{lin2025design,
  author = {Hong-Ye Lin and Xuemei Bai and Chun Li and Shuguang Xu and Haibin Xu and Zne-Jung Lee and Yun Lin and Qing Zhou and Jinhai Cai},
  title = {{Design and Application of an Energy Management System Based on Artificial Intelligence Technology}},
  year = {2025},
  doi = {10.3390/engproc2025091016},
}
Figure