Prediction model of measurement errors in current transformers based on deep learning.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate long-term monitoring stability of electronic current transformers is vital for power grid current signal acquisition.
• Distinguishing non-stationary signal fluctuations from transformer errors is a significant challenge in power system monitoring.

Purpose of the Study:

• To develop an advanced current transformer error prediction model to enhance monitoring stability and facilitate early fault detection.
• To improve the accuracy and reliability of error evaluation in electronic current transformers.

Main Methods:

• A hybrid deep learning model, CNN-MHA-BiLSTM, integrating Convolutional Neural Networks (CNNs), Multi-Head Attention (MHA), and Bidirectional Long Short-Term Memory (BiLSTM) networks.
• Optimization of BiLSTM model parameters (hidden layer nodes, training frequency, learning rate) using the Golden Jackal Optimization (GJO) algorithm.
• Application of CNN for mining instantaneous error data features and BiLSTM for extracting historical error patterns.

Main Results:

• The proposed CNN-MHA-BiLSTM model demonstrated significant advantages in accuracy and stability for current transformer error prediction.
• Validation on substation transformer operation data confirmed the model's effectiveness in both single-step and multi-step prediction scenarios.
• The integration of GJO and MHA mechanisms enhanced the model's ability to capture subtle data characteristic changes, improving prediction accuracy.

Conclusions:

• The developed GJO-optimized CNN-MHA-BiLSTM model offers a robust solution for accurate current transformer error prediction.
• This model can be widely applied to assess transformer operational stability and enable early detection of potential faults.
• The findings highlight the potential of advanced AI techniques in improving the reliability and safety of power grid infrastructure.