A Novel Transformers Fault Diagnosis Method Based on Probabilistic Neural Network and Bio-Inspired Optimizer

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Computational Intelligence

Background:

• Traditional dissolved gas analysis (DGA) methods for transformer fault diagnosis lack sufficient accuracy and stability for modern engineering demands.
• Developing robust and accurate fault diagnosis systems is crucial for power grid reliability.

Purpose of the Study:

• To propose a novel artificial intelligence-based fault diagnosis method for transformers.
• To enhance the diagnostic accuracy and stability of probabilistic neural networks (PNN) using a bio-inspired optimizer.

Main Methods:

• A probabilistic neural network (PNN) was employed as the core classifier.
• An improved salp swarm algorithm (ISSA), incorporating sine cosine algorithm and disruption operator, was used to optimize the PNN's hidden layer smoothing factor.
• The ISSA-PNN model was validated using real-world sensor data and compared against traditional and machine learning methods.

Main Results:

• The ISSA-PNN model demonstrated superior learning ability for complex fault data.
• The proposed method showed significant advantages in accuracy and robustness compared to support vector machine (SVM), back propagation neural network (BPNN), multi-layer perceptron (MLP), and the international electrotechnical commission (IEC) ratio method.
• The ISSA effectively improved the exploration capability and convergence speed of the salp swarm algorithm.

Conclusions:

• The developed ISSA-PNN method offers a more accurate and stable solution for transformer fault diagnosis.
• This artificial intelligence approach provides a promising alternative to conventional methods, particularly for complex fault scenarios.
• The optimization strategy enhances the performance of PNN, making it more suitable for practical engineering applications.