Bio-Inspired PHM Model for Diagnostics of Faults in Power Transformers Using Dissolved Gas-in-Oil Data

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Prognostics and Health Management (PHM) systems are crucial for equipment efficiency and maintenance cost reduction.
• PHM systems rely heavily on component data for accurate fault diagnosis.
• Transformer faults in power grids pose significant risks, necessitating advanced diagnostic techniques.

Purpose of the Study:

• To propose a novel bio-inspired PHM model for diagnosing transformer faults in power grids.
• To optimize the Back-propagation Neural Network (BPNN) structure using the Bat Algorithm (BA).
• To enhance the BA with a chaos strategy for improved optimization and avoidance of local optima.

Main Methods:

• A modified Bat Algorithm (MBA) incorporating chaos strategy for improved initialization.
• Optimization of Back-propagation Neural Network (BPNN) using the Bat Algorithm (BA).
• Utilizing a dissolved gas-in-oil dataset (DGA) for transformer fault diagnosis.

Main Results:

• The proposed Bat-BPNN model achieved a fault diagnosis accuracy of 97.14%.
• This represents a significant increase from the baseline accuracy of 77.14%.
• The MBA-BPNN model demonstrated superior performance compared to BPNN, PSO-BPNN, and GA-BPNN.

Conclusions:

• The developed bio-inspired PHM model offers enhanced fault diagnostic performance for power grid transformers.
• The modified Bat Algorithm effectively optimizes neural networks, leading to higher accuracy.
• This approach provides a promising solution for improving the reliability and efficiency of power grid infrastructure.