Time Series Data Augmentation for Energy Consumption Data Based on Improved TimeGAN.

Area of Science:

• Energy Management
• Artificial Intelligence
• Manufacturing Systems

Background:

• Accurate time series energy consumption prediction is crucial for optimizing manufacturing efficiency and reducing operational costs.
• Deep learning models for sensor data prediction are effective but highly dependent on data quantity and quality.
• Limited data availability in real-world manufacturing settings hinders model performance.

Purpose of the Study:

• To improve the accuracy of time series energy consumption prediction models in manufacturing environments.
• To address the challenge of limited training data by employing data augmentation techniques.
• To enhance deep learning models for manufacturing energy usage forecasting.

Main Methods:

• Utilized an improved TimeGAN model with a multi-head self-attention mechanism for energy consumption data augmentation.
• Developed a hybrid Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) model for predicting manufacturing equipment energy consumption.
• Evaluated model performance using metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²).

Main Results:

• Data augmentation using the improved TimeGAN significantly enhanced the prediction accuracy of the hybrid CNN-GRU model.
• Observed substantial reductions in RMSE and MAE, alongside an increase in the R² value after data augmentation.
• Optimal prediction accuracy was achieved when the synthetic data volume was approximately double the original dataset size.

Conclusions:

• The proposed data augmentation strategy effectively overcomes limitations of sparse time series data in manufacturing.
• The improved TimeGAN combined with a hybrid CNN-GRU model offers a robust solution for accurate energy consumption prediction.
• Findings suggest that augmenting data to twice the original size maximizes prediction model performance in this context.