Research on Intelligent Matching Technology of Power Grid Dispatching Automation Emergency Plan Based on Large Model and Small Sample LoRA Fine-Tuning Algorithm

In the context of large-scale new power system access, the scale of the dispatch automation system is expanding, and it becomes extremely challenging for operation and maintenance workers to achieve efficient and accurate matching between fault descriptions and contingency plans in the face of a large amount of complicated data. In order to solve this problem and speed up the fault handling process, this paper proposes a fault profile matching algorithm for power automation systems based on a large language model and low-rank adaptive fine-tuning. The method reduces the consumption of memory and explicit memory for the training and prediction of the large language model by low-rank coding, and uses the language comprehension capability of the large language model to improve the accuracy and response speed of the automatic matching of fault phenomena and fault scenarios. Improving the understanding and recognition of keywords in the power domain by a large language model by freezing the weights of a large pretrained model and incorporating the power dispatch automation system fault profile information into the pre-training layer of the original Transformer architecture. This method significantly reduces the number of trainable parameters for the downstream task while retaining the text comprehension and querying capabilities of the original model, and demonstrates better performance advantages and accuracy for the field of fault profile matching in power dispatch automation.