Intelligent Verification of Relay Protection Settings

Intelligent verification of relay protection settings combines deep learning, real-time simulation, and automated text analysis to ensure accurate, efficient, and reliable substation protection.

Deep Learning-Based Verification

Modern intelligent substations leverage deep learning models to automate the verification of relay protection settings. An improved convolutional recurrent neural network (CRNN) can identify relay setting text from images, converting them into feature sequences using CNNs and analyzing them with RNNs. Enhancements like bidirectional convert gate long short-term memory (Bi-CGLSTM) allow adaptive weighting of data, improving recognition accuracy. Verification is then performed using Chinese word segmentation, a dictionary of standard setting names, the Levenshtein distance algorithm for similarity calculation, and an improved forward maximum matching algorithm to match and verify each setting individually. Experimental results show identification accuracy exceeding 97% and verification accuracy of 97.07%, with faster processing than traditional methods .

Real-Time Simulation Verification

Closed-loop real-time simulation is another critical method for verifying relay settings. Unlike static secondary injection tests, real-time simulation evaluates relay performance under dynamic fault conditions, including source shifts, breaker status changes, and stressed coordination scenarios. This approach ensures that relays respond correctly to actual system behavior, verifying timing, logic, and I/O interactions before energizing the substation. It reduces the risk of commissioning errors that could lead to unnecessary outages or misoperation .

Practical Maintenance and Verification Cycles

Routine verification and maintenance remain essential for reliable operation. Relay protection devices should be inspected to ensure correct wiring, intact circuits, and accurate settings. Verification cycles depend on system voltage and reliability requirements: high-reliability or 60kV+ systems typically require annual verification, while 10kV systems may be calibrated every two years. Regular verification ensures selectivity, sensitivity, rapidity, and reliability of the protection system, preventing the escalation of faults and maintaining safe, economic operation of the power grid .

Integrated Approach

Combining deep learning-based text recognition, real-time simulation, and structured maintenance cycles provides a comprehensive framework for intelligent verification. This approach enhances accuracy, efficiency, and operational safety, supporting the intelligent operation and maintenance of modern substations while minimizing human error and verification time .