Fuzzy reinforcement learning based intelligent classifier for power transformer faults.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems

Background:

• Existing power transformer fault classifiers exhibit low identification accuracy and fail to detect all fault types.
• Accurate fault detection is crucial for preventing catastrophic failures and ensuring grid stability.

Purpose of the Study:

• To propose an adaptive, intelligent fuzzy reinforcement learning (RL) based classifier for detecting incipient power transformer faults.
• To achieve high on-line fault identification accuracy for all transformer fault types.

Main Methods:

• Utilized dissolved gas analysis (DGA) data from real power transformers as input.
• Employed the J48 algorithm to select the 8 most relevant input variables from 24 DGA variables.
• Developed a fuzzy RL based classifier for adaptive, on-line fault identification.

Main Results:

• The proposed fuzzy RL classifier achieved an exceptional fault identification accuracy of 99.7%.
• This accuracy significantly surpasses that of other contemporary soft computing-based fault identifiers.
• Experimental results demonstrate the superiority and efficacy of the fuzzy RL technique.

Conclusions:

• The fuzzy reinforcement learning approach offers a highly accurate and effective solution for power transformer incipient fault classification.
• This intelligent classifier addresses the limitations of previous methods, providing reliable condition monitoring.
• The technique shows significant promise for enhancing the operational safety and reliability of power transformers.