Small Sample Building Energy Consumption Prediction Using Contrastive Transformer Networks.

Area of Science:

• Energy Management
• Artificial Intelligence
• Machine Learning

Background:

• Accurate energy consumption prediction in large exposition centers is challenging due to data scarcity and fluctuating usage patterns.
• Existing methods often struggle with limited datasets, hindering effective energy management strategies.

Purpose of the Study:

• To introduce a novel algorithm, the contrastive transformer network (CTN), for predicting energy consumption in large exposition centers.
• To address the challenges of limited datasets and fluctuating electricity usage patterns in energy prediction models.

Main Methods:

• The study utilizes a contrastive transformer network (CTN) based on self-supervised learning.
• Contrastive learning is applied across temporal and contextual dimensions.
• A transformer-based architecture is employed for efficient feature extraction.

Main Results:

• The CTN demonstrates strong performance in predicting energy consumption, particularly in scenarios with limited data samples.
• Experiments on a proprietary dataset validate the effectiveness of the CTN algorithm.
• The network excels at capturing complex energy usage patterns in expansive structures.

Conclusions:

• The contrastive transformer network (CTN) offers a potent solution for energy consumption prediction in large exposition centers.
• The proposed method shows significant promise for improving energy management and efficiency in such facilities.
• Self-supervised learning combined with transformer architecture is effective for energy prediction with scarce data.