An Auxiliary Method of Relay Protection Setting for Transmission Lines Based on LLaMa2-7B

The precise operation of relay protection devices for transmission lines is contingent upon the accuracy of setting values. The current approach to the operation and maintenance of relay protection setting values lacks sufficient intelligence, predominantly depending on specialized soft-ware. In response to these challenges, this study introduces an auxiliary method for relay protection setting based on LLaMa2-7B. The proposed model not only provides setting values for relay protection but also offers recommendations based on transmission line data. Initially, a dataset comprising 252 relay protection setting lists for 110kV-220kV transmission lines from a regional grid was assembled and preprocessed. Subsequently, the LLaMa2-7B model was modified and fine-tuned for training, enabling it to focus on the task related to protection setting values prediction while maintaining general language generation capabilities. Cross-validation and task separation techniques are employed to enhance predictive accuracy. Finally, case results indicate that the model's root mean square error in predicting setting values, with line parameters and control words as input, is as low as 0.45%, which is a 12.83% reduction compared to the direct fine-tuning method using the dataset. This proposed method facilitates the precise prediction of relay protection set-ting values, thereby aiding in the validation and calibration process.