Substation equipment temperature prediction based on multivariate information fusion and deep learning network

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Substation equipment temperature prediction is complex due to seasonality, instability, and limited data.
• Traditional methods struggle with the dynamic and multivariate nature of thermal data.

Purpose of the Study:

• To develop an accurate and robust method for predicting substation equipment temperature.
• To leverage multivariate information fusion and deep learning for enhanced thermal analysis.

Main Methods:

• Feature engineering using correlation analysis to create a multivariate information fusion feature vector (MIFFV).
• Dimensionality reduction of MIFFV using Principal Component Analysis (PCA) to obtain a reduced feature vector (RFV).
• Deep learning models, including Convolutional Neural Networks (CNN) for feature extraction and Gated Recurrent Units (GRU) for time-series prediction.

Main Results:

• MIFFV incorporating ambient, time, and space features significantly improved prediction performance over single or dual-feature vectors.
• The proposed CNN-GRU model, using RFV as input, demonstrated superior prediction accuracy compared to four other models.
• Evaluation metrics included Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE).

Conclusions:

• The proposed multivariate information fusion and deep learning approach effectively addresses the challenges in substation equipment temperature prediction.
• The integration of CNN and GRU provides a powerful framework for analyzing complex thermal dynamics.
• This method offers a promising solution for improving the reliability and maintenance of electrical infrastructure.